00:00:00
This guy is literally gonna hover over
00:00:02
us.
00:00:04
This pilot hates podcasting. What a
00:00:07
prick.
00:00:09
>> The besties are broadcasting from the
00:00:11
USA House at the World Economic Forum.
00:00:14
Our episode is sponsored by the New York
00:00:16
Stock Exchange. Are you looking to
00:00:18
change the world and raise capital? Do
00:00:20
it at the NYSC. The NYSE is a modern
00:00:24
marketplace and a massive platform built
00:00:26
for scale and long-term impact. So, if
00:00:29
you're building for the future, the NYSC
00:00:31
is where it happens.
00:00:38
>> I'm Jason Calakanis. This is the All-In
00:00:39
interview show. Uh, last minute we got
00:00:42
added to the roster here at the World
00:00:45
Economic Forum, and we had time to do a
00:00:49
halfozen interviews and, uh, Brian was
00:00:52
here, and uh, this is your Brian
00:00:55
Armstrong from Coinbase, of course, and
00:00:57
friend of the pod. This is probably your
00:00:58
fourth or fifth appearance on the pod.
00:01:00
You come to Davos because this actually
00:01:03
isn't for you about networking. This is
00:01:05
about serious regulations on a global
00:01:09
basis. Yeah.
00:01:10
>> Well, that's been the focus of this
00:01:12
attendance at Davos is we are trying to
00:01:14
get market structure legislation done
00:01:15
for crypto. But actually there I mean
00:01:17
there is a lot of networking that
00:01:19
happens here. We've done a lot of
00:01:20
commercial meetings. You know five of
00:01:22
the top 20 global banks are now using
00:01:24
Coinbase to build their crypto
00:01:25
infrastructure into their products. uh
00:01:27
we meet with heads of leaders of
00:01:29
different countries and talk to them
00:01:30
about economic freedom and how crypto
00:01:32
can update their financial system. So
00:01:33
there's all kinds of good meetings.
00:01:35
>> So uh you had embedded in there these
00:01:38
partnerships with banks. Uh is that like
00:01:40
a white label type thing so they can
00:01:42
sell crypto to their customers? Is it
00:01:44
disclosed which banks are doing that and
00:01:45
how that works?
00:01:46
>> Um a couple of them are public. Uh we've
00:01:48
talked about integration with JP Morgan,
00:01:52
PNC Bank. Um, you know, there's a couple
00:01:56
others that are not public yet, but five
00:01:57
of the top 20 gibs are now using
00:02:00
Coinbase for that. And then we're also
00:02:01
powering integrations with like Black
00:02:03
Rockck and, you know, they they've said
00:02:04
they want to tokenize every single one
00:02:06
of their funds. And so, a lot of these
00:02:08
financial institutions are coming on
00:02:09
chain, which is great.
00:02:10
>> And this is quite I mean, I was thinking
00:02:13
on the way over here how you've really
00:02:16
struggled to work with regulators over
00:02:18
the last decade. I remember under the
00:02:21
Biden administration, the 46th
00:02:23
administration, you went to DC and were
00:02:25
like, "I'm here. I would love to talk to
00:02:26
you." And they were like, "Yeah, we
00:02:28
don't want to talk to you."
00:02:30
>> Now, some people might have varying uh
00:02:34
feelings about Donald J. Trump, our our
00:02:36
47th president. But one thing he has
00:02:39
nailed is interfacing with the business
00:02:42
community and taking regulation and
00:02:45
creating a legal path for crypto
00:02:47
specifically very seriously. What's the
00:02:49
how's the last year? How have the how
00:02:51
have things changed for you in the last
00:02:53
year?
00:02:53
>> Yeah. Well, I know you like to call
00:02:54
balls and strikes and I think just
00:02:56
looking at it objectively, you know, the
00:02:58
Biden administration really tried to
00:03:00
unlawfully kill this industry in America
00:03:02
from my point of view. And Donald J.
00:03:04
Trump, you got to give him credit. I
00:03:05
mean, he campaigned on this idea of
00:03:07
making the United States the crypto
00:03:09
capital of the world. He's kept his
00:03:11
promises. He's leaned in and tried to
00:03:13
get clear rules and regulations passed
00:03:14
so that uh American companies can
00:03:17
thrive, American consumers can earn more
00:03:19
money on their money. Um and he
00:03:20
understands also it's an important
00:03:22
political issue. There's a huge base of
00:03:24
there's like 52 million Americans who've
00:03:25
used crypto now,
00:03:26
>> right?
00:03:26
>> And they want to see clear rules. They
00:03:28
want to see this, you know, get better
00:03:30
financial services in the United States.
00:03:32
It's also, frankly, a global
00:03:34
competitiveness issue, right? I mean,
00:03:35
China just announced that they're going
00:03:36
to pay interest on their central bank
00:03:38
digital currency. Some of the largest
00:03:40
stable coin issuers are still offshore.
00:03:43
He wants to repatriate that capital and
00:03:45
bring it into the US.
00:03:46
>> This is the crazy thing we went through.
00:03:49
I was never a fan calling balls and
00:03:52
strikes of people doing things that
00:03:54
weren't buttoned up.
00:03:55
>> Mhm.
00:03:55
>> But I was even less of a fan of the
00:04:00
prior administration just not meeting
00:04:02
and saying, "Hey, this is uniquely
00:04:04
different. let's figure out a way to
00:04:06
give you a path to do it properly. And
00:04:09
so,
00:04:10
>> you know, in our industry, sometimes you
00:04:13
have to reinterpret rules. Airbnb
00:04:16
uh Uber the biggest success of my
00:04:17
investment career like they bent rules
00:04:19
too. Crypto bent some rules. Um some
00:04:22
cases people broke them uh and uh they
00:04:24
paid the price. But here we are now the
00:04:26
rule set is being refined. The most
00:04:28
important one I think for you is stable
00:04:31
coins and your competition with the
00:04:34
bank. You have banks as partners, but
00:04:36
you're also a competitor to them. Yeah.
00:04:40
>> Well, I I'd say it's mostly
00:04:41
collaborative. I'd say the of the bank
00:04:43
CEOs that I've met with here, I most of
00:04:45
them are actually very into crypto.
00:04:47
They're they're starting to integrate
00:04:48
it. You know, I met with one of the top
00:04:49
10 global banks in the world yesterday
00:04:52
and the CEO told me, "Crypto is is my
00:04:54
number one priority. We view that this
00:04:55
is existential. We're all in. We're
00:04:58
going to put all
00:04:59
>> Why is it existential for them? What do
00:05:01
you think?" they're seeing it's like
00:05:03
it's like when the internet came around
00:05:05
um you know and you had Amazon competing
00:05:06
with Barnes & Noble or you had blogs
00:05:09
competing with not New York Times like
00:05:11
in print, right? Yes.
00:05:13
>> And so anytime there's always change
00:05:15
happening in the world and you can think
00:05:16
of it as an opportunity or you can think
00:05:18
of it as a threat and bury your head in
00:05:19
the sand and pretend it's not happening.
00:05:21
But the reality is that uh crypto is
00:05:24
massive like something like 500 million
00:05:25
people have used it globally. You know
00:05:27
Bitcoin was the best performing asset
00:05:28
class of the last decade. um the largest
00:05:31
financial institutions of the world are
00:05:32
now integrating this. And so at this
00:05:34
point, I don't I think it's foolish to
00:05:36
pretend that this isn't happening. And
00:05:37
we also, by the way, have the Genius
00:05:38
Act, the stablecoin bill is now passed
00:05:40
into law. So we're not going to undo
00:05:43
that. That is that is law of the land.
00:05:45
Like Congress just put that into law.
00:05:47
>> And it's very important because
00:05:50
um what uh David Saxs uh my my bestie um
00:05:55
I think led there was these have to be
00:05:58
audited. These have to be above board.
00:06:01
We can't have a run on stable coins,
00:06:03
which let's face it, people anticipated
00:06:05
Tether would have at some point. There
00:06:07
were lots of fines they got. There was
00:06:10
these attestations people didn't know if
00:06:11
they even had the resources they had.
00:06:13
And now it's pretty clear you have to
00:06:15
keep your assets in treasuries. Correct.
00:06:18
That's correct. Under the Genius Act
00:06:19
that got passed into law last year, US
00:06:21
regulated stable coins have to have 100%
00:06:23
of the assets stored in short-term US
00:06:26
treasuries. So I something like 30 days.
00:06:29
30-day treasuries are the max I believe.
00:06:31
So um that's pretty much the safest
00:06:33
thing you can get. You know, you're
00:06:35
basically trusting the United States
00:06:36
government is not going to fail in 30
00:06:38
days, which I think is a pretty safe
00:06:39
bet.
00:06:40
>> I'm going to go safe bet.
00:06:41
>> Yeah. And um you know, I've been making
00:06:43
this point as well that um you know,
00:06:45
banks do something called fractional
00:06:47
reserve lending. They actually don't
00:06:49
store all your money there. They're
00:06:50
lending it out. That's why they have
00:06:51
such high regulatory overhead because
00:06:54
there can be a run on the bank and it
00:06:56
gives them a very unique business model.
00:06:58
Um they can basically lend it out. You
00:07:00
know the the old joke is like you lend
00:07:02
it out at 6% um you pay three and you're
00:07:05
on the you know the tea time by 3:00 or
00:07:08
whatever. But um that business model is
00:07:10
not available to you unless you have a
00:07:11
business a bank license. But in a stable
00:07:15
coin world with 100% reserves, you don't
00:07:16
need a bank license for that and and you
00:07:18
can give people
00:07:19
>> because it's safer,
00:07:21
>> right? And we saw this Silicon Valley
00:07:23
Bank essentially
00:07:25
>> had mistimed their allocations with um
00:07:30
>> uh treasuries, I guess. And what
00:07:33
happened? They had a run on the bank.
00:07:34
Literally, I was in a board meeting and
00:07:37
in the board meeting on I think it was
00:07:38
like a Thursday and the run happened
00:07:40
Thursday afternoon.
00:07:41
>> I get a text like get your money out of
00:07:43
Silicon Valley Bank. I'm in the board
00:07:45
meeting. We're having it's on the docket
00:07:47
like the third thing is to talk about
00:07:48
Silicon Valley Bank and we're have 100%
00:07:50
of our money in there. Two of the board
00:07:51
members are like we can't just take all
00:07:53
the money out of Silicon Valley. They've
00:07:54
been incredible partners for 30 years. I
00:07:55
said how about we take out half so we
00:07:57
can make payroll. I insisted.
00:07:59
>> Yeah.
00:07:59
>> Literally that night, boom. Uh and so
00:08:03
the key issue now is your business
00:08:05
model. You you need to have revenue and
00:08:09
the revenue from these stable coins is
00:08:12
paying some interest and the people who
00:08:15
are putting their money in there being
00:08:16
able to make some interest on their
00:08:18
hard-earned capital. Yeah, that's the
00:08:20
sticking point for you.
00:08:22
>> Yeah. And it's not interest, it's a
00:08:23
rewards program. This was carefully
00:08:25
negotiated in the Genius Act.
00:08:27
>> And yes, that's that's our view. I mean,
00:08:29
look, in my opinion, it's actually
00:08:31
>> What's the difference there? What what
00:08:32
the reward program, we should think of
00:08:34
it like American Express points.
00:08:36
>> Yeah. I mean, there's lots credit card
00:08:38
reward programs, but the difference
00:08:39
legally is that rewards can't be based
00:08:42
solely on the balance you're holding.
00:08:43
The customer has to do some sort of
00:08:44
other activity like payments or trading
00:08:46
or they have a subscription to Coinbase
00:08:48
1. So, when customers do that, we
00:08:51
actually pass along about 100% of the
00:08:53
economics to them uh for holding those
00:08:56
stable coins with us. And that's a big
00:08:58
driver of growth. Now, you know, there's
00:09:01
always been um this balance between do
00:09:04
people want to put their money in money
00:09:06
markets or do they want to put it as
00:09:07
bank deposits. Um I think that that is
00:09:10
not this crypto is not really new in
00:09:12
that dimension. Like it's just another
00:09:14
flavor of this happening. And um there's
00:09:16
been a lot of hand ringing about, you
00:09:18
know, this is going to destroy all the
00:09:19
lending market. And like I don't think
00:09:20
that's true. Like money markets are
00:09:22
already trillions of dollars and there's
00:09:23
high depos high yield checking accounts.
00:09:25
But these banks haven't had to deal with
00:09:26
a disruptive competitor, a technology
00:09:29
competitor who's really good at what
00:09:31
they're doing. So, they're a little bit
00:09:32
nervous about their franchise. Yeah.
00:09:34
>> Is my interpretation. Am I correct?
00:09:36
>> Some of again, some of them are nervous,
00:09:37
some of them are leaning into it as an
00:09:38
opportunity. I think the latter we want
00:09:40
everyone to win here. I think that's
00:09:41
what um you know, I don't speak for the
00:09:43
president, but like my interpretation of
00:09:45
his comments is that he wants all
00:09:47
American businesses to win. There there
00:09:49
is a win-win outcome here. Um, but if
00:09:51
someone, you know, is going to try to
00:09:52
undermine his legislation that just got
00:09:55
passed in Genius, he'd probably
00:09:57
>> Is that what's happening now? The banks
00:09:58
are trying to retrade the deal.
00:10:00
>> I, you know, I don't want to like I
00:10:02
mean, be careful here. I There's the
00:10:03
bank trade groups which I believe um are
00:10:06
trying to undo the Genius Act. It just
00:10:09
got passed into law 4 months ago. And
00:10:11
for us, that's a red line. I've talked
00:10:13
to many others in the industry that for
00:10:14
them that's a red line. I think we have
00:10:16
to accept that that's law and that's
00:10:18
going to continue to exist. But that
00:10:19
doesn't mean banks and crypto companies
00:10:21
can both win in this new world.
00:10:23
>> Yeah. So this is just a classic tale of
00:10:26
uh incumbents.
00:10:27
>> Yeah. Incumbents and and new folks and
00:10:30
you want to partner with them. You want
00:10:31
to enable it? You have a good
00:10:33
partnership with Jeremy Lair, old friend
00:10:35
of mine at Circle. Are they like the
00:10:37
default stable coin in Coinbase or how
00:10:40
do you think about the relationship with
00:10:41
them? How should we think about the
00:10:42
relationship with them? Yes, we've got a
00:10:44
strong relationship with Circle and USDC
00:10:46
is the largest regulated stable coin
00:10:48
because it is compliant under Genius in
00:10:50
the US. They're compliant under Mika in
00:10:52
Europe, etc., etc. Um, you know, there's
00:10:55
another one that you're familiar with
00:10:56
which is still,
00:10:57
>> but I think it's in the process of being
00:10:59
>> they're trying to clean it up is my
00:11:01
understanding.
00:11:02
>> Yeah.
00:11:02
>> So that they can participate. The likely
00:11:05
scenario is there'll be two tethers, a
00:11:07
United States one and then that complies
00:11:10
and then there'll be the Wild West one
00:11:12
outside the US. Is that what you've
00:11:13
heard as well?
00:11:14
>> Yeah.
00:11:14
>> Yeah.
00:11:15
>> Yeah. And I should mention um we don't
00:11:16
have like an exclusive with Circle or
00:11:18
anything like that. We we actually list
00:11:19
other stable coins on our platform.
00:11:20
>> Do you list Tether?
00:11:21
>> We support it in certain ways. Um you
00:11:24
know, especially people who want to
00:11:25
convert Tether, but we we support it. We
00:11:27
also support PayPal stable coin. We're
00:11:29
open to listing others, too. So, we we
00:11:31
don't have an exclusive on USDC.
00:11:32
>> Yeah. But you're not endorsing it. And
00:11:34
do you let people trade into Tether or
00:11:37
just let them trade out of Tether? Like,
00:11:38
how does it work mechanically?
00:11:40
>> I think it's different in different
00:11:41
countries. I want to make sure I get it
00:11:42
exactly right. But in countries where
00:11:44
we're allowed to do it, I mean, we
00:11:45
support Tether,
00:11:46
>> right?
00:11:47
>> It's it's nuanced.
00:11:48
>> Are you concerned about or have you been
00:11:51
historically concerned about Tether and
00:11:53
like their little bit of a loosey goosey
00:11:56
approach to regulations and trading? I'm
00:11:59
giving it that descriptor, not you. But
00:12:01
I I would think having it on your
00:12:04
platform with regulators pretty focused
00:12:06
on it over the last 5 or 10 years and
00:12:08
this, you know, belief that this thing
00:12:11
could all come apart and create a run
00:12:13
that to you is just not worth the risk
00:12:15
to to be too close to it in case it does
00:12:17
flip over. Yeah.
00:12:19
>> Yeah. We we've definitely gotten
00:12:20
questions on it. And look, I I want to
00:12:21
be careful here. I actually like the
00:12:22
Tether guys. I think they've done a lot
00:12:24
of good things in the world. There's
00:12:25
people who really are they're struggling
00:12:27
with local currencies that are have like
00:12:29
70 100% inflation year-over-year. And so
00:12:31
there was high demand for the dollar.
00:12:33
They got great distribution in a lot of
00:12:34
the emerging markets. I actually think
00:12:36
they've done a lot of good for the
00:12:37
world. Um but yeah, it's not currently
00:12:40
compliant under the Genius Act in the
00:12:42
US. And so it doesn't follow those same
00:12:44
100% reserves and short-term US
00:12:45
treasuries is my understanding. So
00:12:47
people have to make their own
00:12:48
determination on that. And I think other
00:12:51
countries are following suit in terms of
00:12:53
cleaning this up. Is there in crypto now
00:12:56
a way to give a rating that is sort of
00:12:59
objective for consumers to say like this
00:13:02
one has this grade?
00:13:04
>> Mhm.
00:13:04
>> And follows these regulations. Hey, this
00:13:06
one is follows this level. It's a lower
00:13:09
grade and this one doesn't follow and
00:13:11
it's you know it's a memecoin. Whatever.
00:13:13
This is the wild west. Like no crying in
00:13:15
the casino coins. like what what is your
00:13:18
responsibility as a platform or
00:13:20
opportunity as a platform to like inform
00:13:22
the people who are participating?
00:13:24
>> Yeah. So what we try to do is have
00:13:28
minimum listing standards. Um so if we
00:13:31
believe that there's a cyber security
00:13:33
risk to it um the developer could you
00:13:35
know rug everyone or if if it's really
00:13:38
it's illegal from a compliance point of
00:13:39
view. You know there's a few different
00:13:41
areas we look at. So if it meets the
00:13:43
minimum bar, we will list it and then we
00:13:44
let customers decide it. I don't feel
00:13:47
like it's our job to be recommending
00:13:49
investments, you know, like in the
00:13:50
traditional financial world, there's
00:13:52
like these AAA rated bonds and you know,
00:13:54
and I it always felt a little bit like
00:13:56
those organizations that do the ratings
00:13:58
are kind of always be politicized and
00:14:00
like
00:14:00
>> go see the big short.
00:14:01
>> Yeah, exactly. So I don't it's I think
00:14:04
of it a little bit like um the app
00:14:06
stores, right? I mean you or the or
00:14:09
let's say Amazon. I mean, you want to
00:14:10
have the everything exchange. You want
00:14:12
to have the everything store. You know,
00:14:14
everything that's legal should be in the
00:14:15
store, but um maybe there's customer
00:14:17
reviews we we could add at some point.
00:14:19
We actually tried that for a little bit.
00:14:20
Like if you see a two out of five star
00:14:21
thing on Amazon, you can still buy it,
00:14:22
but it's kind of at least you're
00:14:24
informed. We we tried making user
00:14:26
ratings at one point. It didn't go that
00:14:27
well cuz people were basically voting
00:14:29
with their whatever they like, you know.
00:14:31
>> Sure. They're talking their book.
00:14:32
>> Yeah. Talking their book. So, um,
00:14:34
anyway, we we right now we're in the
00:14:36
regime of just disclosures and minimum
00:14:38
listing standards.
00:14:39
>> Yeah. Which crypto projects do you find
00:14:42
the most fascinating right now? The Bit
00:14:45
Tensor one, some of these ones that are
00:14:47
popping up that are actually providing
00:14:50
technological solutions to problems,
00:14:52
distributed computing, etc. You I find
00:14:54
those fascinating.
00:14:54
>> They are. I mean, people are trying to
00:14:56
tokenize data centers and like oil
00:14:58
reserves. I think I think the biggest
00:15:00
trends happening in crypto right now is
00:15:03
number one it's the everything exchange.
00:15:05
So it's not just crypto you can trade um
00:15:07
you're getting equities you know are
00:15:08
increasingly getting closer to being
00:15:10
able to trade uh onchain you know um
00:15:14
prediction markets are
00:15:15
>> you have a partner for that right?
00:15:16
>> Yeah we're working with KI currently
00:15:19
>> is that exclusive or are you willing to
00:15:21
put anybody up on the thing?
00:15:22
>> It's not exclusive so we're looking at
00:15:24
others as well I know you guys work with
00:15:26
Poly Market or whatever on the show. I
00:15:27
mean, I'm a I was one of the original
00:15:29
angel investors in uh Robin Hood and I
00:15:32
think they're doing Kelsey, too. It
00:15:34
seems like people are plugging different
00:15:35
ones in. Poly Market's our favorite.
00:15:37
Yeah.
00:15:37
>> Yeah. We So, we're talking to Poly
00:15:38
Market, but you know, we can also list
00:15:40
our own um prediction markets.
00:15:42
>> Oh, yeah. So, you could fire up your
00:15:43
own.
00:15:43
>> Yeah. So, anyway, we're along I think
00:15:45
that Yeah, the biggest trends are all
00:15:47
assets are coming on chain for trading.
00:15:49
Um prediction markets are growing like
00:15:50
crazy and stable coin payments are
00:15:52
growing like crazy. Those are the
00:15:53
probably the three biggest trends in
00:15:54
crypto right now. When do you think
00:15:55
stable coins tip into the area of
00:15:57
businesses
00:15:59
two-part question businesses using it
00:16:01
for payments? You know, maybe
00:16:04
>> uh to reduce friction. You know, some
00:16:06
designer does some work for Coinbase and
00:16:08
makes sure a new logo and you want to
00:16:10
send them $25,000. It goes through a
00:16:12
stable coin. When does that start to
00:16:14
happen? And then for consumers, when do
00:16:15
people at a poker game start settling up
00:16:17
with, you know, through their Coinbase
00:16:19
account with a stable coin? Yeah. So the
00:16:21
biggest growth area over the last year
00:16:22
has been B2B crossber payments and
00:16:25
crossboarder.
00:16:26
>> Yeah. Cross border especially because
00:16:27
there's a lot of these companies you
00:16:28
know they might be buying goods from
00:16:30
Asia or Europe trying to sell it in
00:16:32
their shop in Brazil or whatever it is
00:16:33
and they have to wait 7 days and there's
00:16:35
high FX fees and all this kind of stuff
00:16:37
to move
00:16:37
>> crazy the fees.
00:16:38
>> Yeah. Uh so that's been growing really
00:16:41
nicely. You know we we launched a
00:16:42
product called Coinbase Business which
00:16:44
is serving lots of small and mediumsized
00:16:46
companies who want to do crossber
00:16:47
payments and invoicing and tax and
00:16:48
accounting and all that. How do you find
00:16:50
those customers? You I mean that that's
00:16:52
like a real unique group of people or do
00:16:53
they just find you?
00:16:54
>> Currently they're beating a path to our
00:16:56
door. We we actually have a huge backlog
00:16:58
we need to people waiting to onboard. So
00:17:00
we need to staff up that team.
00:17:02
>> Um we also launched something called
00:17:04
Coinbase developer platform uh which
00:17:06
provides it's kind of like AWS if you
00:17:08
want. So that's like you can white label
00:17:09
anything like with the banks but lots of
00:17:11
other businesses are using that for
00:17:12
wallets, trading, payments, staking,
00:17:15
financing, all kinds of things. You
00:17:16
mentioned uh tokenization. I was talking
00:17:19
with Vlad. He did a little experiment.
00:17:22
Hey, why don't I tokenize some OpenAI
00:17:23
shares? Sam wasn't Sam Wman wasn't too
00:17:26
thrilled with that. How do you think
00:17:28
about that opportunity? I'm a private
00:17:30
market investor. I would love to be able
00:17:32
to take my early position in Robin Hood
00:17:35
or my early position in Uber as it was
00:17:38
going up and put it into a market and
00:17:40
let people trade it.
00:17:41
that would be very interesting uh to for
00:17:43
VCs for you know angel investors to be
00:17:46
able to move that around. How do you
00:17:47
think about it?
00:17:48
>> Well, I think it has to be done with the
00:17:49
permission of the companies cuz you know
00:17:51
if you're if you're a private company
00:17:52
you don't want your employees to be able
00:17:54
to get liquid after one year. you're
00:17:55
trying, you know, that's why you have
00:17:56
vesting. That's why it's it's a
00:17:58
retention mechanism, right? Let's all
00:17:59
build this together and like maybe when
00:18:01
we go public because there are stories
00:18:03
of founders who like took a little
00:18:05
secondary too early, then the company
00:18:06
didn't didn't work out and it's, you
00:18:08
know, bad off.
00:18:09
>> So, I think what's going to happen in in
00:18:11
crypto is that um in the private markets
00:18:14
like first of all, we should make
00:18:16
onchain capital formation way easier for
00:18:18
private companies. So, if you want to go
00:18:21
um this is this is what we're chatting
00:18:22
with the SEC about and others is like,
00:18:24
you know, can you go register a
00:18:25
security? Um right now you'd only be
00:18:28
able to raise money from credit
00:18:29
investors in the US. I know you you and
00:18:30
I agree on this like we'd like to expand
00:18:32
how you can become an accredited
00:18:34
investor.
00:18:34
>> And there is a bill right now that's
00:18:36
working its way and it would basically
00:18:38
mean the SEC and and they've already
00:18:40
been charged with this, but they they I
00:18:42
don't know if you've uh studied the SEC
00:18:43
at all. They tend to take their time and
00:18:45
then not do what they've been told to
00:18:47
do. Uh, and that was one of the things
00:18:49
they were supposed to do is create an
00:18:50
accreditation test.
00:18:52
>> Uh, well, this I say this SEC is
00:18:54
actually moving very quickly. But this
00:18:55
one is Yes.
00:18:56
>> Yeah. Um, but yeah, I think that would
00:18:58
be a more fair way because otherwise
00:19:00
it's kind of like a regressive tax like
00:19:02
only rich people can get richer on
00:19:03
private investments. But um, anyway, I
00:19:05
think onchain capital formation is going
00:19:07
to be massive for private companies. I
00:19:09
think eventually like you'll actually
00:19:11
just be able to go public totally
00:19:12
onchain too. And um yeah, these markets
00:19:15
are just going to get
00:19:16
>> would lower the cost massively, reduce
00:19:18
the friction and increase the
00:19:20
democratization of wealth creation.
00:19:23
>> Yeah.
00:19:23
>> If you think about when you were a
00:19:25
private company.
00:19:26
>> Mhm.
00:19:27
>> Like you had all this pent-up demand,
00:19:30
people trying to buy the shares like
00:19:31
crazy. They were all doing backdoor kind
00:19:33
of shady stuff, popping up SPVS. I don't
00:19:35
know if you've been following the SPV
00:19:37
market now, but there's like it's turned
00:19:39
into a boiler room. It's no longer Chris
00:19:41
Saka
00:19:42
>> representing Twitter and you know doing
00:19:44
an orderly thing or
00:19:46
>> Elon doing every six months an orderly
00:19:48
thing where SpaceX keeps control of it.
00:19:50
Now people are going out raising money
00:19:52
from dentists and civilians high net
00:19:53
worth individuals to buy SpaceX or
00:19:56
Andrea or whatever it is and then they
00:19:57
go try to find the shares and they
00:19:59
charge them 10% loadin fee no carry.
00:20:02
>> Mhm.
00:20:02
>> I mean think about how crazy that is.
00:20:04
>> Yeah. there's such high demand for some
00:20:06
of these large private companies and
00:20:09
>> you know it's kind of a good example of
00:20:10
like the unintended consequences of
00:20:12
higher regulation sometimes.
00:20:13
>> Yeah.
00:20:14
>> Like um you know Sarbain Oxley and all
00:20:16
that kind of stuff really cut down the
00:20:18
number of how how long companies stayed
00:20:19
private before they went public and then
00:20:21
yeah I mean Uber and Airbnb and a lot of
00:20:23
these things like they all the money was
00:20:25
made by private investor or credit
00:20:26
investors like you know yourself and
00:20:28
then when they went finally went public
00:20:29
it kind of went sideways.
00:20:30
>> Oh yeah. Airbnb, Uber, they all had like
00:20:32
a 5-year uh indigestion period, I would
00:20:35
say, you know, and it it and some
00:20:37
people, I think Instacart wound up going
00:20:39
from 30 billion down to 10 billion when
00:20:41
they went public and it was like, okay,
00:20:42
we got to dig out of a hole the last
00:20:44
series of investors. And that is the
00:20:45
unintended consequence of this because
00:20:47
you don't have anybody setting a proper
00:20:49
valuation for the company in some
00:20:51
reasonable way. What about funds? You
00:20:53
know, I get approached by a lot of
00:20:55
people offshore, etc., hey, take your
00:20:57
next seed fund. You're going to do a $50
00:20:58
million fund. to put it on the chain.
00:21:00
And then, hey, if you were one of my LPs
00:21:02
and you were like, "Oh, I need
00:21:04
liquidity. I could just sell it to
00:21:05
somebody else or if I was an LP in a
00:21:07
Sequoia fund and I wanted to sell you
00:21:09
the interest, you could buy it for me
00:21:10
and we could just
00:21:11
>> take our wallets out and zip zip."
00:21:13
>> Yeah, I think that's that's absolutely
00:21:15
going to happen. I mean, Coinbase
00:21:16
launched a product actually called
00:21:17
Coinbase tokenize. So, we're helping any
00:21:19
fund or real estate project or anybody
00:21:21
who wants to tokenize their products and
00:21:23
it just democratizes access. It
00:21:25
increases demand. It gets rid of a lot
00:21:27
of the back back office fees, gets rid
00:21:29
of the settlement risk because it can be
00:21:30
settled instantly on chain and there's
00:21:32
some very innovative company. I mean
00:21:34
like Black Rockck, Apollo, like these
00:21:35
the top funds in the world are putting
00:21:37
they they've come out publicly and said
00:21:38
they want to tokenize every single one
00:21:39
of their products. It's it's absolutely
00:21:41
happening.
00:21:42
>> How do they keep control of it? Because
00:21:44
you have this more liquid back to
00:21:47
consequences or this would be downstream
00:21:49
effects, second order effects, third
00:21:51
order effects. What are the third second
00:21:53
and third order effects that would
00:21:55
happen if a venture fund or a REIT were
00:21:57
onchain?
00:21:58
>> Have you thought that through?
00:22:00
>> Have you thought it through? Yeah.
00:22:01
>> Yeah. I mean there so there's different
00:22:02
types of funds. Um some are going to be
00:22:04
only available to institutions
00:22:05
accredited. Some would be open to retail
00:22:07
and so um for the retail side I mean you
00:22:10
could actually get I don't know like
00:22:12
tens of millions of people around the
00:22:14
world in five minutes to all put in in
00:22:16
the average price might be $100 or
00:22:18
$1,000, right? So it starts to really
00:22:20
democratize access. I mean, we actually
00:22:21
just published this um this report
00:22:24
recently and people have heard about the
00:22:25
unbanked, but there's actually 4 billion
00:22:28
adults who are unbrokered as well, which
00:22:29
means they don't have any ability to
00:22:30
invest in these highquality assets. So,
00:22:32
it's like this is the engine of wealth
00:22:34
creation for capitalists, you know, like
00:22:36
you and I. A lot of people are just
00:22:38
stuck. The only way they can earn is
00:22:39
from their labor, right? and they they
00:22:42
want to probably put like even if
00:22:44
they're they have $100 or $1,000, they
00:22:46
might want to put 10% of that into the
00:22:48
S&P 500 or Coinbase stock or whatever,
00:22:50
Nvidia, whatever. And they can't do that
00:22:52
because
00:22:52
>> but they can use prize picks. They could
00:22:54
use some other thing. No, and I like
00:22:56
prize picks. I use prize picks to bet on
00:22:57
Nick's parlays. Uh but they they wind up
00:23:01
putting it somewhere else and why not
00:23:02
being able to, you know, if they want to
00:23:04
bet on the Knicks, that's fine. But or
00:23:06
go to Vegas, that's fine, too, and play
00:23:07
in a poker tournament. Yeah, maybe they
00:23:08
hear about this company LinkedIn because
00:23:10
they work in the HR department. Yeah.
00:23:12
And everybody in the HR department's
00:23:14
over the moon about it. That rank and
00:23:16
file 75K person working in HR, they
00:23:20
understand what the next big product
00:23:23
will be. They'll know Indeed or
00:23:25
LinkedIn's going to work and they can
00:23:26
make a life-changing bet with just
00:23:28
$1,000. They make it a,000x return.
00:23:30
>> Yeah. Yeah. I mean, there's a financial
00:23:32
literacy component to this as well. And
00:23:33
I think actually the AI agents are now
00:23:36
getting really good. that we've
00:23:36
integrated one into Coinbase app. It can
00:23:38
kind of teach people about dollar cost
00:23:40
averaging and you know tax loss
00:23:41
harvesting. So there's a the financial
00:23:43
educa literacy is is a part of it but
00:23:45
then yeah let's make high quality
00:23:47
investments available to them and
00:23:50
democra it's just like lifting people
00:23:52
out of poverty. It's great. I mean
00:23:53
you're thinking about the whole globe
00:23:55
but just thinking in the United States a
00:23:57
lot of what people are upset about and
00:23:58
the topic we've been talking about a lot
00:24:00
um is the rise of socialism in
00:24:03
>> um New York specifically California. I'm
00:24:06
not sure if you're still a resident. I
00:24:07
won't put you on the spot here, but
00:24:09
>> considering options
00:24:10
>> considering I mean
00:24:11
>> as we all are.
00:24:12
>> I two years ago I moved to three years
00:24:14
ago I moved to Austin. I was like I'm
00:24:16
done. I just the social issues and you
00:24:18
know I saw the writing on the wall. I
00:24:19
mean I actually think there's a chance
00:24:21
>> that the that California goes bankrupt
00:24:23
and I said that on the podcast like this
00:24:24
feels like it's trending towards
00:24:26
insolveny. I never thought they would
00:24:27
get to the wealth tax or just seizing
00:24:29
people's assets.
00:24:30
>> Yeah.
00:24:30
>> What's your take on all that?
00:24:33
Some I was reading something this
00:24:34
morning which said that actually just
00:24:36
the people who have already left which
00:24:38
is probably like
00:24:39
>> my my estimates would probably be 20% of
00:24:41
the billionaires have already left in
00:24:42
California
00:24:43
>> that it's already created a negative 10
00:24:45
billion tax hole even if even with the
00:24:47
amount that they hope to raise from the
00:24:49
people who stay.
00:24:50
>> Yeah.
00:24:50
>> So it's one of the biggest cell phones
00:24:52
I've ever seen. Um it's a disaster and I
00:24:55
uh you know I'm torn actually because
00:24:58
part there's always kind of this
00:24:59
question of voice or exit right like
00:25:01
voice do you try to fix it from within
00:25:03
exit do you leave like you did
00:25:05
>> and I think um
00:25:06
>> the incentives are strange because on
00:25:08
the one hand like I love California um
00:25:10
but on the other hand it's been kind of
00:25:12
um it's like an abusive relationship you
00:25:14
know it just keeps coming back with
00:25:15
another thing and another thing
00:25:16
>> and in some ways like if you do make the
00:25:18
decision to leave um you don't have
00:25:20
really much incentive to try to fix it
00:25:22
at that point you actually want to get a
00:25:23
lot of the the builders and the top
00:25:26
talent out of California at that point
00:25:27
and just all resettle in a new place
00:25:29
that is welcoming to us and businesses.
00:25:32
>> Yeah. I don't think people understand
00:25:34
how easy it is for somebody who's in a
00:25:37
certain strata to is already operating
00:25:40
globally. I'm on planes and on four
00:25:42
different continents every year, you
00:25:44
know, like I it doesn't matter where I
00:25:47
am. It matters my wife and my kids are
00:25:49
happy and they love the place we live
00:25:50
and we love Austin. And then the thing
00:25:52
I've seen you the probably one of the
00:25:54
hardest thing you had to deal with with
00:25:56
your employees and your team at Coinbase
00:25:58
is the price of their housing.
00:25:59
>> Like how many times did you try to
00:26:01
recruit somebody to come to California
00:26:03
>> and it's like a family of, you know,
00:26:05
four and they need private school, they
00:26:07
need a house and you're like, "Oh my
00:26:08
god, what is their nut going to be
00:26:10
here?"
00:26:10
>> Yeah. My their nut. I love that. There's
00:26:14
a great South Park episode on that. But
00:26:16
um yeah, I mean I you're right. That's a
00:26:18
major barrier whenever we make an offer
00:26:20
to somebody um in California, you know,
00:26:22
or for New Limit the Biotech, like
00:26:24
that's always like, well, my cost of
00:26:25
living is going to double like you need
00:26:26
to pay me more. So, it's getting
00:26:28
expensing through to you, the business
00:26:30
owner.
00:26:31
>> Yeah, it does. I mean, San Francisco had
00:26:33
this um kind of like revenue based tax
00:26:36
that was very punitive on um financial
00:26:39
services companies. Stripe Stripe moved
00:26:40
out when that happened.
00:26:42
>> I think Mark Benoff regrets supporting
00:26:44
that one.
00:26:44
>> Yeah, I think so.
00:26:45
>> Yeah. But in fairness, he did want to
00:26:49
finance the homeless industrial complex,
00:26:51
which has been completely ineffective in
00:26:53
reducing the number of homeless
00:26:55
individuals because
00:26:56
>> they're not homeless. They're addicted
00:26:58
to drugs.
00:27:00
>> A home doesn't help that problem.
00:27:01
>> Exactly. I mean, I I'm I'm preaching to
00:27:03
the choir here, but yeah, I think people
00:27:05
would have a lot more tolerance to pay
00:27:07
higher taxes if they felt like it was
00:27:09
working. But the history of the last 10
00:27:11
years in California is that the budget
00:27:13
has gone up dramatically and the
00:27:15
services have gotten worse. It's like
00:27:16
it's actually creating the wrong
00:27:18
incentives. Like the more ser the more
00:27:21
money we spend on homelessness, the more
00:27:22
homeless people are. So, you know, and
00:27:24
then the waste and fraud. Oh my gosh.
00:27:25
Like you guys Nick Shirley and all that.
00:27:28
I think
00:27:29
>> what do you what I mean what could we
00:27:31
even guess is the level of abuse in
00:27:33
California? It's going to it's going to
00:27:35
make Minnesota look
00:27:37
>> like peanuts. Such a big economy with so
00:27:40
many NOS's and so many homeless
00:27:42
organizations taking down hundreds of
00:27:44
millions of dollars in San Francisco
00:27:46
alone.
00:27:47
>> Yeah.
00:27:48
>> Um so tell everybody about the the side
00:27:52
hustle your your other company. Yeah.
00:27:53
That you
00:27:54
>> Oh, the biotech.
00:27:55
>> Yeah.
00:27:55
>> Yeah. Well, when Coinbase went public in
00:27:58
um 2021, you know, I got some liquidity
00:28:00
from that and I just I thought it
00:28:03
through and I was like, "All right, I
00:28:04
want to be CEO of continue to be CEO of
00:28:05
Coinbase. Being a public company CEO is
00:28:07
a really cool thing. I just feel like
00:28:08
we're at the beginning of our journey."
00:28:10
>> Um but I also felt like I wanted to
00:28:11
start to use some of that capital to go
00:28:13
after these like big bets, right? I was
00:28:15
kind of a little inspired by Elon
00:28:16
actually. I think, you know, he did the
00:28:18
thing with um PayPal and X and um and
00:28:21
then, you know, he went into these like
00:28:23
the world of atoms, not bits, right?
00:28:25
It's actually like in software is more
00:28:27
forgiving. Startups are all hard, but
00:28:28
software is a little more forgiving
00:28:30
because you have higher margins. The
00:28:31
world of atoms is much less forgiving.
00:28:33
So anyway, I was lucky enough to meet um
00:28:35
some really amazing co-founders um that
00:28:38
came together with this idea in the
00:28:40
longevity space and it's called, you
00:28:41
know, there's the fundamental science
00:28:42
behind it is called epigenetic
00:28:43
reprogramming. It's sort of you can
00:28:45
reprogram your cells to restore function
00:28:47
they had when they were younger. There
00:28:48
was some really cool research being
00:28:49
done. I hosted a couple dinners. Anyway,
00:28:51
I decided to uh fund these guys and a
00:28:54
bunch of other people have invested now
00:28:55
and uh I'm a board member. I've been
00:28:57
helping them in in small
00:28:59
>> and the name of it is
00:29:00
>> New Limit. Yes,
00:29:00
>> New Limits.
00:29:01
>> So, they've been incredible progress.
00:29:03
It's
00:29:03
>> when will they have a product? This
00:29:04
feels like a 20-year investment, not a
00:29:07
two or five.
00:29:08
>> Yeah. Well, biotech does move more
00:29:10
slowly, but um it's moved faster than I
00:29:12
would have thought. I thought this was
00:29:13
going to be like five years of just
00:29:15
basic research. M um but it turned out
00:29:17
actually within the first uh 2 or 3
00:29:19
years they were able to successfully
00:29:21
demonstrate reprogramming of human cells
00:29:23
to restore function they had when they
00:29:24
were younger and the first drug
00:29:26
candidate is probably going to go into
00:29:27
clinical trials next year. So
00:29:29
>> amazing.
00:29:29
>> Yeah,
00:29:30
>> that's super rewarding. Five minutes.
00:29:32
Yeah. Okay, great. Um so
00:29:37
coming out of Davos, uh what's your take
00:29:40
on the state of the world? Everybody
00:29:43
when they get here seems to uh it seems
00:29:45
like the ESG DEI kumbaya stuff
00:29:49
>> has switched in the last year or two to
00:29:53
you know brass tax deal making whether
00:29:55
it's between countries
00:29:57
um and businesses like this is turning
00:29:59
into a business conference that used to
00:30:01
be this is what everybody's telling me
00:30:03
on the streets in in the you know in the
00:30:05
houses
00:30:07
>> um it's it's about business now and on
00:30:09
the margins there's a patina of you know
00:30:12
globalization versus nationalism. What's
00:30:14
your take on the state of the world in
00:30:17
2026 talking to regulators, talking to,
00:30:20
you know, people who work in government?
00:30:22
>> Mhm.
00:30:23
>> I think you're right. I mean, I've only
00:30:24
been at Davos once before, but it did
00:30:27
feel more um, you know, how do we make a
00:30:29
global government? How do we do lots of
00:30:31
ESG and DEI? And that's really not what
00:30:35
anyone's talking about now. I think
00:30:36
partially it's because of Larry Fing
00:30:37
coming in, the new leader defect, you
00:30:39
know, more or less. And um I also think
00:30:42
it's because of Donald Trump. I mean,
00:30:43
>> yeah, he shook it up.
00:30:44
>> Yeah. Like the numbers that the United
00:30:46
States is putting up in terms of GDP
00:30:48
growth and low inflation and just that
00:30:50
business environment. It's like, hey,
00:30:52
how do we all win? That's how you create
00:30:53
prosperity for everyone in society. Um I
00:30:55
do think it actually benefits everyone
00:30:57
like you know even the poorest people in
00:30:58
society are they do the best in high
00:31:01
economic freedom countries that are
00:31:02
anyway.
00:31:03
>> Growth solves a lot of problems.
00:31:04
>> It does. It does. I mean, and the growth
00:31:06
is objectively calling balls and strikes
00:31:10
spectacular. We have not seen this level
00:31:12
of GDP since we pumped a bunch of money
00:31:14
and printed a bunch of dollars during co
00:31:16
like
00:31:17
>> 5.6 GDP is pretty spectacular. Let's
00:31:19
hope it keeps up. Unemployment very
00:31:22
reasonable 4.6 lowest of our lifetime. I
00:31:24
think 4.3 was the lowest it hit.
00:31:26
Inflation, yeah, closer to three than
00:31:27
two, but you know, the actual average
00:31:30
has been like 2.8 2.9. So the two is the
00:31:32
target.
00:31:33
>> Yeah. So we're right around the average.
00:31:34
not hitting the target yet, but I think
00:31:36
we'll get there. So, yeah, it seems like
00:31:38
>> growth does not come from government
00:31:40
spending, right? That's like the key
00:31:42
that this is this Keynesian economic
00:31:44
argument I think is basically wrong.
00:31:45
Like growth comes from uh having like
00:31:48
deregulation, having lowcost energy,
00:31:51
allow the private markets to bill, let
00:31:53
them have clear rules about what's
00:31:55
allowed and not and then create a level
00:31:56
playing field. Everyone competes, the
00:31:58
consumer benefits, the b the companies
00:32:00
benefit, all the employees, the
00:32:01
shareholders. Capitalism is like the
00:32:03
biggest win-win. You know that someone
00:32:05
had a great rant about that recently.
00:32:06
>> Yeah.
00:32:06
>> And so we're seeing the private
00:32:08
companies in the United States really
00:32:10
cook. It's great,
00:32:11
>> right? And if you're cooking, uh, you
00:32:13
create more jobs, hopefully pay more
00:32:15
taxes, all that just starts the cycle in
00:32:17
the right direction. How do you I'll end
00:32:19
on AI. Big debate on the pod. You
00:32:21
haven't been on with the four of us in a
00:32:22
while, but when somebody gets sick,
00:32:24
we'll definitely rotate you in.
00:32:25
Everybody loves when you are on like the
00:32:27
the quartet. You're like a fan favorite,
00:32:29
by the way. Um, but I'm curious what you
00:32:32
think about AI and job displacement. Sax
00:32:34
and I have been debating this. When is
00:32:35
it going to be here? Is it here? Young
00:32:37
people can't find jobs, but we're still
00:32:39
a pretty low unemployment rate overall,
00:32:42
but then Elon's position and Bernie
00:32:44
Sanders position is in sync. Hey,
00:32:46
listen. It's going to be a lot of job
00:32:48
displacement. So, how do you think about
00:32:50
it? Obviously, Amazon also is the one
00:32:52
I'm watching because the idea that
00:32:54
somebody's going to drive packages or
00:32:55
pack packages in the age of Optimus and
00:32:58
Robo Taxi and Whimo sounds crazy.
00:33:01
>> So, those jobs are going away. How do
00:33:02
you think about job displacement? And
00:33:04
what are you seeing with the most AI
00:33:06
first employees in Coinbase?
00:33:09
>> Yeah. Well, just zooming out for a
00:33:11
second, I think like crypto and AI are
00:33:12
the two most important technology trends
00:33:13
happening in the world. And what's cool,
00:33:15
most people don't realize actually
00:33:16
they're going to come together because
00:33:18
AI agents need to get work done and they
00:33:19
have to do payments
00:33:21
>> and the whole traditional financial
00:33:22
system is built around kind of knowing
00:33:24
your a human behind every product with
00:33:26
your you upload your
00:33:27
>> Oh, know your customer. Yeah.
00:33:28
>> Yeah. So AI agents I think are going to
00:33:30
use stable coins and wallet crypto
00:33:32
wallets.
00:33:32
>> Know your agent.
00:33:33
>> Well, I don't even know if you need to
00:33:35
know the just but yeah. Anyway, that's
00:33:37
one of the important trends that we're
00:33:38
trying to help happen in terms of job
00:33:40
displacement. Um I don't know. I I maybe
00:33:43
this is a bit of a techno optimist take
00:33:45
um but I actually think you know if you
00:33:47
go back and look at like the 19 early
00:33:48
1900s I think it was like 80% of the US
00:33:51
population was in was working in
00:33:53
agriculture
00:33:54
>> and so that's like hard manual labor out
00:33:56
in the fields and when agriculture got
00:33:59
automated you know now it's like 3% or
00:34:01
something of the workforce working
00:34:03
>> that happened over 30 years. Yeah.
00:34:04
>> Yeah. And so I think they would look at
00:34:07
what you and I do for a living like
00:34:09
we're just having a cool conversation in
00:34:10
Davos talking. They'd be like that's not
00:34:12
a real job. Like real job is like manual
00:34:14
labor in the fields, right?
00:34:15
>> Uh you just you're on vacation all the
00:34:17
time. But we we think of it as a job and
00:34:18
you know people who are like typing on a
00:34:20
keyboard all but you get to sit in an
00:34:21
air conditioned office or whatever.
00:34:22
>> It's stressful. That's for sure.
00:34:24
>> It can be stressful but that's just I'd
00:34:26
rather be doing that than be doing
00:34:28
backbreaking labor in this in the sun
00:34:30
digging a ditch or something. Right. So
00:34:32
I think that job displacement is like
00:34:34
not a bad thing actually. Um if it means
00:34:37
that people can do new kinds of work and
00:34:39
new kinds of job. Now is there going to
00:34:40
be a transition period? Yes. I I
00:34:43
basically I think if the AI uh plays out
00:34:45
as we all think it will with robots and
00:34:47
a lot of there will be job displacement
00:34:50
but it means that we'll be in a world of
00:34:52
more abundance and people are going to
00:34:53
have jobs that's there like they're
00:34:55
streaming on video games on YouTube or
00:34:57
whatever. Like I don't know what it's
00:34:58
going to be but
00:34:59
>> or they'll think about the future.
00:35:00
they'll think about like great philosoph
00:35:02
you know philosophical works might be
00:35:04
written uh because we don't have to
00:35:06
burden ourselves with like the tedium of
00:35:09
packing boxes. So I I'm basically an
00:35:11
optimist on it. I think it'll be good.
00:35:12
>> I'm pretty optimist about it as well. I
00:35:14
just having watched the robo taxi
00:35:18
self-driving thing
00:35:20
>> and just watching the velocity that it's
00:35:22
getting better and having watched the
00:35:23
Uber story up close for 12 years. I'm
00:35:26
like yeah that's going to happen in six.
00:35:28
Mhm.
00:35:28
>> I think it's just like it's going to
00:35:30
just
00:35:31
>> Yeah. It feels to me and the thing I'm
00:35:33
starting to see in the field is you have
00:35:35
in Wuhan there and and Beijing, they're
00:35:38
having protests and they're saying,
00:35:39
"Well, we're just going to give out a
00:35:40
certain number of self-driving license.
00:35:42
We're going to contain it."
00:35:43
>> And then you have Boston and a couple
00:35:45
and California now. They're saying,
00:35:46
"Hey, listen. We're only going to allow
00:35:48
a certain number of robotoxies." Or
00:35:49
Boston's like, "We're not going to let
00:35:50
you have them here. We're going to
00:35:51
protect these jobs." So, this is going
00:35:53
to become, I think, one of these uh very
00:35:58
class, you know, debates we're going to
00:36:00
have over the coming years. But what
00:36:01
about employees in the company? You have
00:36:03
the same number of employees as you had
00:36:04
a couple years ago. You overhired for a
00:36:06
bit maybe or hiring for growth.
00:36:09
>> Mhm.
00:36:09
>> Now, how do you think about hiring
00:36:11
versus hiring and training young people
00:36:13
versus just automating stuff or just
00:36:16
having those AI? You must have some
00:36:17
people on the staff who are using claude
00:36:19
co-work or something and they're just
00:36:21
like 10x knowledge workers. Not
00:36:24
developers are obvious, but yeah, talk
00:36:26
to me about knowledge workers and what
00:36:27
you're seeing with your most AI first
00:36:28
employees.
00:36:30
>> Yeah. So, one of the big pushes we made
00:36:32
in the last year was um we got our own
00:36:35
internal hosted AI model that was
00:36:38
connected to all of our data sources,
00:36:39
right? So, it's like every Slack
00:36:40
message, every Google doc, every
00:36:42
Salesforce data confluence, you know. So
00:36:45
now um this is all linked up in one like
00:36:48
the data is all aggregated and you can
00:36:50
ask these agents. So every team's legal
00:36:52
is sorry every team is using it um legal
00:36:54
finance everything just
00:36:55
>> it's like the oracle of Coinbase.
00:36:57
>> Yeah. And I've started to ask it really
00:36:59
it's not just like prompting it hey can
00:37:01
you write this kind of memo for me or
00:37:03
something. It's like I'm asking these AI
00:37:05
agents now um as CEO like what should I
00:37:08
be aware of um in the company that I
00:37:10
might not be aware of? and it'll tell
00:37:11
me, "Did you know that like there's
00:37:13
actually disagreement on this team about
00:37:14
the strategy?" And I was like,
00:37:15
"Actually, I didn't know that cuz it can
00:37:17
read every Slack message in every every
00:37:19
Google doc." And then, you know, I've
00:37:21
been prompting it like I actually Toby
00:37:23
on my board, he he he said this is like
00:37:25
it's call he's calling it reverse from
00:37:27
Shopify.
00:37:27
>> Yeah. Yeah. He he said this is like
00:37:29
reverse prompting. So, instead of
00:37:30
telling the AI agent what you want to
00:37:32
do, you ask it what you should be
00:37:34
thinking more about,
00:37:35
>> right? And
00:37:36
>> it's a mentor.
00:37:37
>> Yeah. It's like a coach.
00:37:38
>> Yeah. Like, what could make me a better
00:37:40
CEO? And it's like, well, I notic I
00:37:41
looked at all how you spent your time in
00:37:43
the last quarter. Here's how you said
00:37:45
that you wanted to spend it, but you
00:37:46
actually spent like 32% of your time on
00:37:48
this instead of 20. Um, and it'll I've
00:37:51
asked it other questions like, you know,
00:37:52
what's the thing that I changed my mind
00:37:54
on the most over the last year? Things
00:37:56
like that. So, it's been it's now
00:37:58
becoming like it'll it'll like prompt
00:38:01
you with information you should be
00:38:02
thinking about instead of the other way
00:38:04
around.
00:38:04
>> I recently did this and I I don't know
00:38:06
if you play with Claude Co yet. Uh, have
00:38:09
you played with it? this week play with
00:38:11
there's claude opus 4.5 or something.
00:38:13
>> Yeah. But there's co-work which is kind
00:38:15
of like
00:38:16
>> you describe what you want to do as a
00:38:18
knowledge worker and it starts to build
00:38:20
it. So instead of doing vibe coding and
00:38:21
saying hey I want to write code for this
00:38:23
you just describe like what do you want
00:38:24
the end application and the result to be
00:38:26
and then it kind of like a wizard kind
00:38:28
of takes you through it. It's pretty
00:38:29
scary cuz I connected my notion to it,
00:38:31
my Slack to it
00:38:32
>> and my Google Docs and it did the same
00:38:34
type of thing. I was like, "Tell me
00:38:35
about myself." And it was like, "Whoa,
00:38:37
you need to spend more time with your
00:38:38
founders that are winning as opposed to
00:38:40
more time with your internal team." It
00:38:41
was
00:38:42
>> really interesting to analyze your uh
00:38:45
your teams. And I think that's going to
00:38:46
be like the future of this. All right.
00:38:48
Listen, Brian, you got a lot more
00:38:49
meetings to do. Uh thanks for coming on
00:38:51
the program.
00:38:52
>> I'm thrilled because my guest started
00:38:55
building AI chips seven six seven years.
00:38:58
Six, seven years before Chat GPT was
00:39:02
launched. Andrew Felman is of course the
00:39:03
CEO of Cerebra Systems and they are
00:39:06
building wafer scale engine WSE. Yep.
00:39:11
>> That's the category of chips you're
00:39:12
working on and they're for inference.
00:39:14
>> For inference or for training both
00:39:16
>> or for training or both. And you have
00:39:17
one with you? I do.
00:39:19
>> So here is a way for scale engine.
00:39:21
Remember usually chips are the size of a
00:39:24
postage stamp.
00:39:25
>> Yeah.
00:39:25
>> And so this is say 56 times larger than
00:39:28
a B200.
00:39:29
>> Wow. And it's a 4 trillion transistor
00:39:32
part. And for AI work, big chips process
00:39:35
more information and they deliver
00:39:37
results in less time. So you can faster
00:39:39
results for your your query.
00:39:42
>> And what does that cost for the typical
00:39:44
system today? And how does it compete
00:39:46
with like the H100 200s?
00:39:48
>> Well, the first one cost us half a
00:39:49
billion to make.
00:39:50
>> Yes.
00:39:51
>> First one I've heard is the most
00:39:52
expensive.
00:39:52
>> Turns out the first one's the kicker,
00:39:54
>> right?
00:39:55
>> Yeah.
00:39:55
>> These come in a system.
00:39:57
>> All right. And we can deliver the system
00:39:59
on premise. Yeah. Or you can use it in
00:40:01
our cloud.
00:40:03
>> On premise they're about a million
00:40:04
million and a half depending on how you
00:40:06
have it configured. And on in the cloud
00:40:09
you can rent it by the token. So by the
00:40:11
million tokens it'll vary by different
00:40:13
models from 50 cents a million tokens to
00:40:16
several dollars per million tokens or or
00:40:18
you can rent it by the month or the
00:40:20
year. How did you know
00:40:23
seven years before Chat GPT was
00:40:25
launched? Or did you that the AI
00:40:28
revolution would be this fast, furious,
00:40:31
you know, unstoppable? I mean, it and
00:40:34
has what's happened in the last two
00:40:36
years even surprised you?
00:40:37
>> For sure. I I think Yeah, I think
00:40:39
anybody except maybe Sam and Ilia.
00:40:42
>> Yeah.
00:40:42
>> Uh who really saw it? M
00:40:44
>> you know we talked to them in 2015 and
00:40:46
and what they were saying is now it was
00:40:49
unbelievable how how right they've been
00:40:51
but I I think what we saw was the rise
00:40:54
of a new computational problem called AI
00:40:57
and it would put new and different
00:40:59
pressure on a processor and we saw this
00:41:02
on the horizon and we said what would
00:41:04
happen if this got giant
00:41:07
>> and we had no idea how big it would get
00:41:08
or how fast it would come but as a
00:41:11
computer architect you try and think
00:41:13
about could I build a machine that's way
00:41:15
faster at this new thing
00:41:16
>> and will there be enough of it to build
00:41:18
a business around? And so we saw AI on
00:41:21
the horizon. We said to ourselves, could
00:41:23
we build a a processor that would be
00:41:27
unique in its performance? Could we
00:41:29
build something not one or two or three
00:41:31
or five times faster, but 20 or 50 times
00:41:34
faster? And we came to believe we could.
00:41:37
We chose an approach that that solved a
00:41:39
problem that had been open in the comput
00:41:41
industry for 75 years. Nobody had ever
00:41:43
built a chip this big. Um, many smart
00:41:46
guys had failed. Uh, and we delivered it
00:41:49
and it's blisteringly fast.
00:41:51
>> What were the first applications? You
00:41:53
know, Nvidia got to perfect their
00:41:56
compute
00:41:57
>> and really their company off of the
00:41:59
backs of video game
00:42:02
>> uh players, then Bitcoin and crypto. It
00:42:05
was almost like there were just a number
00:42:08
of way points before AI emerged. You
00:42:12
didn't have those.
00:42:12
>> We didn't have that. And you know, if
00:42:14
you look at Nvidia's stock price from
00:42:15
2004 to 2010, it was flat.
00:42:18
>> Yeah.
00:42:19
>> Right. And um they were trying to find a
00:42:22
new market, right? That they had a lot
00:42:23
of the graphics market and that market
00:42:25
was sort of flat. They they found love
00:42:27
with gamers. Um they tried to go into
00:42:29
the supercomput world. Um we were
00:42:33
focused entirely on AI. And so at first
00:42:36
we we found love with with the national
00:42:39
labs, with the military, uh with some
00:42:43
pharma. Um
00:42:45
>> what were the applications they were
00:42:46
using?
00:42:47
>> They were training various types of
00:42:49
models.
00:42:50
>> Got it. And this is before large
00:42:51
language models.
00:42:52
>> This is before language models existed.
00:42:54
>> So they were doing models for sequencing
00:42:58
you know proteins.
00:43:00
>> Sequencing models. They they were doing
00:43:02
different forms of vision models. they
00:43:05
were doing uh work at the at the edges
00:43:07
of high performance computing and AI.
00:43:11
>> What is now um taking the most compute?
00:43:16
We see a lot of applications now images,
00:43:20
video production, the training of the
00:43:22
models, um and really deep learning,
00:43:26
deep thinking, I guess, where it's
00:43:27
firing off many many threaded jobs.
00:43:31
Which one of those is the most compute,
00:43:33
the most limited? Right now,
00:43:35
>> deep research is deep research uses an
00:43:38
enormous amount of compute.
00:43:40
>> Explain to the audience what happens
00:43:42
when they do one of those deep research
00:43:44
queries. As an example, I've been
00:43:46
playing with the latest Claude and they
00:43:48
have um a co-working Yep.
00:43:51
>> co-pilot type application and I made a
00:43:53
prompt every time we have a guest on the
00:43:55
podcast
00:43:56
>> and I had it do, you know, maybe 15 or
00:43:59
20 steps every podcast you've been on,
00:44:02
every news item, a timeline,
00:44:04
>> right?
00:44:04
>> And it was unbelievable when it makes
00:44:08
this document. It's better than anything
00:44:09
a human has ever made for me. And I've
00:44:11
been doing interviews for 20 years. lots
00:44:13
of pretty smart assistants whose job
00:44:15
that exact thing was.
00:44:16
>> And when I tell you they would ask for
00:44:18
48 hours to do a dossier that was 20% of
00:44:22
what this does in under 10 minutes, I
00:44:24
I'm I'm not even joking. And and last
00:44:27
year I told them like you can use it to
00:44:30
kind of get ideas and get some links,
00:44:31
but you know, keep doing it the old way.
00:44:33
So it kind of cut their time from 16
00:44:35
hours to 8. Now it's 16 hours to I don't
00:44:38
need them.
00:44:39
>> Literally don't need them to do this
00:44:40
work. Right. Now imagine if you could
00:44:42
get it in 10 seconds.
00:44:43
>> Yeah.
00:44:43
>> Right. That's what we do.
00:44:45
>> Yeah.
00:44:45
>> That's it. Exactly. And so what what
00:44:47
happens remember we we make AI with
00:44:49
training and we use AI with inference.
00:44:53
All right. And that's the simplest way.
00:44:54
The reason inference is going through
00:44:55
the roof is because everybody's using
00:44:56
AI. Yes.
00:44:58
>> All right. A task that you kicked off,
00:45:01
right? It starts a bunch of little
00:45:02
threads and each of those asks queries
00:45:05
and each of those queries deliver
00:45:07
results that are the input to other
00:45:08
queries. So you've got a cascade of
00:45:11
queries that is going on.
00:45:13
>> It's wild.
00:45:14
>> And each of those requires more compute.
00:45:18
And so you've got, you know, 20
00:45:20
different queries being kicked off. Each
00:45:22
query asks 20 queries. Each one of those
00:45:24
requires 10 or 15 or 20 seconds to get
00:45:27
done in traditional comput. So you you
00:45:29
have this giant waterfall of time and
00:45:32
answers. And we built this part so you
00:45:35
can get all those answers back in 4
00:45:37
seconds in 10 seconds. And when does
00:45:39
that happen? You know, right now it
00:45:40
seems like when I do these kind of deep
00:45:42
research,
00:45:43
>> it's, you know, grab a cup of coffee
00:45:45
time,
00:45:45
>> right?
00:45:46
>> 5 minutes,
00:45:46
>> right?
00:45:47
>> Not 15, but it seems like about 5
00:45:49
minutes is what it averages. When you're
00:45:51
doing an image or a 5-second video, it
00:45:54
seems like it's 90 seconds or so.
00:45:56
>> When does that come down to, you know,
00:45:58
the experience we had with dialup going
00:46:00
to,
00:46:01
>> you know, fiber?
00:46:02
>> That that's the the perfect analogy,
00:46:04
right? When when the internet was slow,
00:46:06
Netflix delivered DVDs and envelopes. I
00:46:09
know you remember this, right? Oh, I do.
00:46:11
When Netflix got fast, when the internet
00:46:14
got fast, Netflix didn't get better at
00:46:15
delivering DVDs. Netflix became a movie
00:46:18
studio.
00:46:19
>> It enabled them to be something
00:46:20
different. It wasn't a change in degree.
00:46:22
It was a fundamental change in kind. And
00:46:24
what speed does for AI is the same. So,
00:46:28
we have customers uh like Cognition who
00:46:32
use us to power their coding engine.
00:46:34
All right. And if you read the tweets
00:46:37
and you read people's comments, they're
00:46:39
they're odd. There is zero latency
00:46:42
between their requests and their
00:46:43
answers. So they can stay in the flow as
00:46:45
they write code. All right. And so this
00:46:47
is the idea. The idea is you shouldn't
00:46:49
have to wait at all. And uh Claude is
00:46:53
not a anthropic is not a customer, but
00:46:55
we recently announced OpenAI,
00:46:57
>> right? They were original investors.
00:46:59
They were
00:47:00
>> and now they've just put in a major
00:47:01
purchase order. they have and this is
00:47:03
really exciting and part of it I think
00:47:05
was because what we could do is we could
00:47:07
deliver extraordinary speed so that the
00:47:09
user experience changed
00:47:11
>> and as we know having watched Google
00:47:13
Larry and Sergey Marissa the team over
00:47:15
there came to a conclusion when we shave
00:47:18
off milliseconds it's the number one way
00:47:20
we get usage to go up
00:47:22
>> that's exactly right published that
00:47:23
paper years ago that said even
00:47:24
milliseconds even even amounts of time
00:47:27
that the individual user doesn't
00:47:29
recognize
00:47:31
noticeable That's exactly what is it GM?
00:47:33
What is the psychological just notible
00:47:35
perception? I believe your mom would
00:47:37
know. She's a behavioral
00:47:38
>> She would know.
00:47:40
>> Just noticeable. It's 15% of whatever
00:47:43
the number is. So like if you could cut
00:47:44
15% off the time people they use it
00:47:47
more, they leave less. Yes. You know,
00:47:49
there's a uh Paul Graham had a had a
00:47:52
great tweet. He said, "I I' I'd use
00:47:54
Google half as much if Chat GPT weren't
00:47:56
so slow."
00:47:57
>> And if you think about that, that's what
00:47:59
happens, right? while you're waiting for
00:48:02
Claude or you're waiting for Chad GPT,
00:48:05
you you get a coffee or you poke around
00:48:07
somewhere else and you've lost the
00:48:10
customer. That's the cost of being slow
00:48:12
is the customer has gone somewhere else
00:48:13
>> or you do what I do which is I have a
00:48:15
nice wide Dell monitor.
00:48:17
>> I have three browser windows open. I pay
00:48:19
for all three services. I have them all
00:48:22
Gemini, I got Claude, I got chat. I pay
00:48:24
for all of them. I'm paying probably
00:48:26
close to $600 700 personally a month.
00:48:29
So, I'm spending 10,000 a year just for
00:48:31
me,
00:48:32
>> right?
00:48:33
>> And I just take the same query and I go
00:48:35
bing bing bing bing and I start them all
00:48:37
>> or start them all. And
00:48:39
>> I'm probably burning like 10 trees. I
00:48:41
mean, there's it's probably probably
00:48:43
being a little greedy.
00:48:44
>> It's it's not 10 trees, but um
00:48:47
>> it uh I I think that's a really
00:48:49
interesting way to manage how slow it
00:48:51
is,
00:48:51
>> right?
00:48:52
>> And so we exist to fix that problem. And
00:48:54
what what we partnered with OpenAI to do
00:48:57
is is to to deliver blisteringly fast
00:48:59
speed across the world's most popular
00:49:01
models.
00:49:02
>> What's the scope of the deal? Uh
00:49:05
>> what we announced was 750 megaww
00:49:08
>> is what was announced.
00:49:10
>> When did we switch from talking about
00:49:13
the number of chips, the number of units
00:49:16
being sold to the amount of power being
00:49:19
sold? It's a little bit confusing for
00:49:20
folks and it started probably about last
00:49:22
summer. So actually what happened was
00:49:24
that the change has been coming a lot
00:49:25
longer. We used to talk about data
00:49:27
centers in terms of square footage. I
00:49:29
got 100,000 square foot data center,
00:49:30
>> right? And now nobody cares how many
00:49:32
square feet they you have. They care
00:49:33
about how much power you have. So the
00:49:35
limiting constraint on data centers is
00:49:37
always their power footprint. And right
00:49:39
now for large deployments, the limiting
00:49:41
constraint is how much power can be
00:49:43
delivered. So by talking about how many
00:49:46
p how much power is delivered, you're
00:49:48
talking in in the unit of the limiting
00:49:50
constraint.
00:49:51
>> Mhm. And so the limiting constraint is
00:49:54
power. We're trying to find this huge
00:49:56
amount of power uh for OpenAI. It'll be
00:49:59
delivered over several years.
00:50:00
>> And you're responsible for the power as
00:50:02
well or is that like a joint effort?
00:50:04
>> So uh it's a cloud uh deal. So
00:50:07
>> Oh, so they're utilizing your cloud. So
00:50:09
you've got to do all the work.
00:50:10
>> We are we are building the the cloud
00:50:12
infrastructure for it.
00:50:13
>> Got it. Where are you going to where are
00:50:14
you building your data centers? What's
00:50:15
the best location here in 2026 to be
00:50:18
placing these things? Is it overn gas?
00:50:20
Is it near hydro? What's the state of
00:50:22
the art now?
00:50:23
>> So the the the cheapest power in the in
00:50:25
the world is is hydro. Yeah.
00:50:27
>> Without question. After that is natural
00:50:29
gas.
00:50:30
>> And so where places have natural gas,
00:50:32
you you have an abundance of relatively
00:50:34
lowcost power,
00:50:35
>> which is Texas.
00:50:36
>> West Texas, Wyoming, um outside the US,
00:50:39
uh in the Caribbean, in Gana, you have a
00:50:41
huge amount of natural gas. You do in uh
00:50:44
you have geothermal, which is its own
00:50:46
thing in in the Nordics. Um, but natural
00:50:49
gas is a is a very inexpensive way,
00:50:52
particularly if it's coming from uh as a
00:50:54
byproduct from from petroleum mining
00:50:57
where where what you have is it used to
00:50:59
be what's called flare off gas. They
00:51:01
used to just throw it away. They used to
00:51:02
just burn it at the top. Bitcoin miners
00:51:04
found the Bitcoin miners found that,
00:51:05
right? And so, we'll take that.
00:51:07
>> That's right. Like, whoa, don't make
00:51:09
that fire make
00:51:10
>> Yeah.
00:51:10
>> So, basically, you look for the existing
00:51:12
flare off. And then tell me about hydro
00:51:14
because it does seem to me that people
00:51:17
knew that for a long time they were
00:51:19
moving data centers there. Heat still an
00:51:21
issue with your chips and others.
00:51:23
>> We're water cooled and and so water is
00:51:25
an extremely efficient coolant.
00:51:27
>> So, you know, we we we knew early on
00:51:30
that we'd be going to water. We were
00:51:31
some of the first production AI systems
00:51:33
to use water. The TPU early on moved to
00:51:35
water as well. Before that, there'd been
00:51:37
some water cooling, mostly in the the
00:51:40
Department of Energy supercomput labs
00:51:42
that they'd used some water.
00:51:43
>> Does it matter because I remember early
00:51:45
on when people were talking about water
00:51:46
to cool 10 years ago, the source of the
00:51:49
water and how cold that water is coming
00:51:50
in or just water's cool enough?
00:51:53
>> No.
00:51:53
>> And you're fine.
00:51:54
>> It depends on on your particular design,
00:51:58
>> but you'd like cooler water.
00:52:00
>> Sure.
00:52:01
>> Is better.
00:52:02
>> So, Alaska and Canada feel pretty good
00:52:03
about that. or or you bring chillers or
00:52:05
or you cool the water,
00:52:07
>> right? You can often you can take
00:52:09
general groundwater or other forms of
00:52:12
other locations for
00:52:13
>> there's a huge misperception today that
00:52:15
this water is not recycled and that AI
00:52:18
is using all this water when it's it's
00:52:20
not even comparable to golf courses.
00:52:22
Let's say
00:52:23
>> first golf courses are are extremely
00:52:25
water inefficient. Most of our data
00:52:27
centers use uh a closed loop,
00:52:30
>> right? So we we're we're passing the
00:52:33
water by the back of our chip. They pull
00:52:36
the heat off. They warm the water. The
00:52:38
warm water goes down and through a
00:52:40
closed loose system is chilled and
00:52:43
pumped back. And so you're not using
00:52:45
>> and the water is not damaged.
00:52:47
>> The water's not damaged. It's just
00:52:48
>> not like some chemicals or anything that
00:52:50
gets put into them.
00:52:51
>> No, not at all.
00:52:52
>> There's a lot of misperceptions about AI
00:52:54
right now. There's a bit of a
00:52:58
it seems like it's almost like there's
00:52:59
some dark PR forces at work trying to
00:53:03
make the data center build out seem
00:53:05
worse than it is. And then there's also
00:53:08
I think maybe some valid concerns around
00:53:10
jobs. When you look at each one of those
00:53:12
issues, what do you think is, you know,
00:53:15
the ones that are most frustrating as an
00:53:17
AI executive building data centers?
00:53:18
>> It's a really good point. I think some
00:53:19
of the hyperscalers sort of they they
00:53:24
made a bad call in the way they they
00:53:25
went into some of these rural
00:53:26
communities. Right. So what are you
00:53:28
looking for? You're looking for a place
00:53:29
that land is cheap that has an abundance
00:53:30
of power and they went to these
00:53:32
communities and they didn't do a good
00:53:34
job talking to people.
00:53:36
>> Right.
00:53:36
>> Right. And
00:53:37
>> tech people didn't do a good job talking
00:53:39
to humans. Right. What a surprise. And
00:53:41
they went into these communities and
00:53:43
they cut deals with the power company.
00:53:45
And the power company was looking to
00:53:48
build new infrastructure to support
00:53:49
them. Yeah.
00:53:50
>> And traditionally the regulated power
00:53:52
industry would then advertise that cost
00:53:55
over 20 or 30 years. So they ended up
00:53:57
increasing
00:53:58
>> the local people's power rates, right?
00:54:02
And so the people got upset and that
00:54:04
very reasonable, very reasonable. If
00:54:05
instead
00:54:06
>> you'd have gone in and said, uh, look,
00:54:09
great. We we we're going to be good
00:54:11
citizens. We're going to be big
00:54:12
taxpayers here. Let's build more
00:54:14
schools. We we can build a school for
00:54:16
you. It's it's a rounding error in the
00:54:18
cost of this facility. We're going to
00:54:21
make a bunch of construction jobs and
00:54:23
we're going to be good citizens. Yes.
00:54:25
>> Um they would have had a very different
00:54:27
approach.
00:54:27
>> And not only that, they kind of were a
00:54:30
little heavy-handed early on where they
00:54:31
said, "We're going to play three
00:54:33
communities off each other and who's
00:54:35
going to give us the biggest tax
00:54:36
discount?" Right.
00:54:37
>> That was another cell phone mistake.
00:54:39
Yeah.
00:54:39
>> Now, what I just saw is that Microsoft
00:54:41
just put out,
00:54:42
>> we talked about it this week on the
00:54:43
show.
00:54:43
>> I thought it was a very thoughtful and
00:54:45
reasonable sort of approach for a
00:54:48
company that's sort of a national
00:54:49
champion, right? That that they're going
00:54:51
to be good citizens. They want to make
00:54:53
sure that that your rates.
00:54:54
>> They basically said just to catch the
00:54:55
audience up, we guarantee you our usage
00:54:57
of energy will not increase the uh
00:55:00
>> cost of your utilities.
00:55:01
>> The cost of your utilities.
00:55:02
>> That's fair. I mean,
00:55:03
>> I mean, very reasonable. And I think the
00:55:06
next step we were brainstorming on the
00:55:07
pod. There are people putting solar on
00:55:10
roofs and uh there's base power. Um
00:55:13
Michael Dell's son doing a really
00:55:14
interesting project.
00:55:15
>> Yeah. Where they just put batteries on
00:55:17
the side of your house. They don't have
00:55:19
to be super intricate and they load
00:55:22
those batteries up when there's extra
00:55:23
power and it's cheap and they deploy it
00:55:25
when you know the duck curve or whatever
00:55:27
the demand hits. So if you think
00:55:28
Microsoft and you talk about rounding
00:55:30
errors, well, if you give everybody a
00:55:31
battery at home to store some energy
00:55:34
when it's cheap and there it's and then
00:55:36
that can be used to flow back into the
00:55:38
data centers. I think we're could live
00:55:39
in a world where you say, "Hey, we're
00:55:41
going to put a data center here and
00:55:42
everybody's energy is free."
00:55:43
>> I think well, there are a couple things.
00:55:45
First is we chose as a nation not to
00:55:47
invest in our grid for 40 or 50 years,
00:55:50
right? And so our our our grid is behind
00:55:52
and vastly in need of improvement,
00:55:54
right? our our grid is is decrepit
00:55:57
compared to other advanced nations and
00:56:00
in particular compared to what what
00:56:01
China has done.
00:56:02
>> I think the ability to store
00:56:06
power at your home and to use it when
00:56:11
power is the the most expensive is an
00:56:13
obviously a reasonable thing to do,
00:56:15
>> right? Obviously the reasonable thing to
00:56:17
do
00:56:18
>> and it takes load off the the grid as
00:56:20
well. And people like the idea of being
00:56:22
a little resilient, right?
00:56:23
>> They do. If you do lose your power,
00:56:24
which in California, I think they turn
00:56:27
it off about on the peninsula, was it
00:56:30
about a half dozen times a year for you?
00:56:31
>> Only when it's really hot or really
00:56:33
cold.
00:56:34
>> Either either one.
00:56:35
>> Yeah.
00:56:36
>> And they'll leave it off for 2 days cuz
00:56:38
it's wind and it's just a complete
00:56:40
disaster. Oh, and by the way, I don't
00:56:41
know if you knew this, there were
00:56:43
subsidies given for nuclear where the
00:56:46
people living around nuclear power
00:56:48
plants in France and where they just
00:56:50
said in exchange for living near a
00:56:52
nuclear power plant, which some people
00:56:55
might have concerns about. Uh maybe
00:56:57
they're reasonable, maybe they're
00:56:57
unreasonable. Put that aside, we're
00:56:59
going to just give you free energy for
00:57:00
life. Very interesting. What is your
00:57:03
thought on small modular and nuclear?
00:57:06
It's just it's too far out. Yeah. For
00:57:08
you to be concerned with right now. I
00:57:09
think it's both the obviously the right
00:57:12
thing to do and probably not the source
00:57:14
of data centers for the next three or
00:57:15
four years.
00:57:16
>> Yeah,
00:57:16
>> both. Right. That that obviously we need
00:57:18
to be working on that. Obviously, it's
00:57:20
an extremely efficient form of power
00:57:23
creation. Um nuclear has uh over 20 or
00:57:28
30 years is vastly more efficient than
00:57:30
any other power we know how to how to
00:57:32
create. Um it has a disadvantage. It
00:57:34
upfront it's a little more expensive,
00:57:36
right? So the fees are are are upfront
00:57:39
and then you get the benefit over years.
00:57:41
But clearly we should be working on
00:57:42
this.
00:57:43
>> And if you had the ability to do one,
00:57:44
you would do it.
00:57:45
>> Oh yeah, for sure. And you're seeing
00:57:47
some of that in the more aggressive
00:57:48
nations. The UAE
00:57:50
>> uh is building uh modular uh nuclear
00:57:53
data center based data centers. Um
00:57:56
putting huge amounts of power on the
00:57:57
grid with nuclear. And what a great
00:58:00
idea.
00:58:00
>> I went to see Elon a couple weeks ago on
00:58:02
a Sunday afternoon. and we're talking he
00:58:05
really thinks um that putting data
00:58:08
centers and chips in space, cooling's
00:58:10
pretty easy in space, solar is much more
00:58:13
effective. What do you think, and he's
00:58:15
been talking about this publicly, so I'm
00:58:16
not speaking out of school, but what do
00:58:18
you think about data centers in space?
00:58:19
Have you started researching it?
00:58:21
>> We have. I think first betting against
00:58:23
Elon's ideas is is probably
00:58:25
>> longterm
00:58:26
>> not a good betting strategy. However, he
00:58:29
getting the timing right is something
00:58:31
that that has been uh less uh a less
00:58:35
>> he's never wrong. He's frequently late.
00:58:37
>> That's right. I always I look I I think
00:58:40
that's the both the blessing and the
00:58:42
curse of visionary.
00:58:43
>> Yeah.
00:58:43
>> Is you see things other people uh can't
00:58:46
see and in your mind they're just some
00:58:48
technical hurdles to overcome.
00:58:50
>> You took a couple years to build that.
00:58:51
>> It took took a couple years.
00:58:52
>> Were you on time?
00:58:53
>> Um
00:58:54
>> we we were we were plus or minus a year.
00:58:56
>> Okay. in delivery of something that
00:58:58
nobody had ever done.
00:59:00
>> Um,
00:59:00
>> it's hard to predict.
00:59:01
>> It's it's it's really hard. I I think
00:59:03
the idea of using uh space for to grab
00:59:08
solar power is obviously a smart idea.
00:59:11
>> Yeah.
00:59:12
>> Right. You're you're you're you're miles
00:59:14
closer to the sun. You have much uh
00:59:17
thinner atmosphere blocking the rays, so
00:59:20
you can gather up the power. I think
00:59:22
there's a lot of technical work to be
00:59:24
done. Yes, it's cold there, but you're
00:59:26
also in a vacuum. So, the actual cooling
00:59:30
isn't an easy problem. It's a solvable
00:59:32
problem. I think the communication among
00:59:35
satellites
00:59:36
>> is a real real issue. And figuring out
00:59:40
which technology you want to do to get
00:59:42
the data back to Earth, right? Remember,
00:59:44
uh when you got your uh your cable TV,
00:59:49
uh when you tried to get internet from
00:59:51
those satellites, it was really glitchy.
00:59:53
Yeah,
00:59:53
>> there were these big delays. Now, those
00:59:55
were higher orbit satellites. The ones
00:59:57
he's thinking about are much lower
00:59:58
orbit. They'd have lower latency, but
01:00:00
there's some real work to be done.
01:00:02
>> I I think it's in the 8 to 10 year
01:00:04
category, not in the 3 to five.
01:00:06
>> Yeah, I I think it maybe split the
01:00:08
difference, but yeah, it's and it it's
01:00:10
all of these are worth pursuing if you
01:00:12
believe that we're not going to
01:00:15
overbuild. So,
01:00:17
knowing what you know and watching this
01:00:19
build out, is it possible that we're
01:00:22
overbuilding right now and we'll need a
01:00:24
digestion period or do you think, you
01:00:28
know, based on what we're seeing there's
01:00:30
just going to be the next workload, next
01:00:32
workload, next workload.
01:00:34
>> I think we are still really early in in
01:00:36
the demand for AI compute. And I I think
01:00:39
if you think about, you know, what
01:00:42
portion of enterprises have really
01:00:46
adopted AI in a meaningful way that has
01:00:49
changed their workflow, it's tiny. I
01:00:52
think even the most frequent users at
01:00:54
the consumer level are going six, eight
01:00:56
times a day. What what when what happens
01:00:58
when they go to 100 times a day? What
01:00:59
happen when all their devices are going
01:01:01
for them? What what happens when
01:01:03
everybody in GNA, right? when every
01:01:07
engineer is using it as a as a coding
01:01:09
co-pilot, right? We're going to see
01:01:11
enormous amounts of demand for
01:01:13
inference. The models are getting
01:01:14
better, more people are using, they're
01:01:16
using more often, and the the amount of
01:01:19
compute taken with each usage is
01:01:22
increasing. So, I think we're just at
01:01:24
the beginning. How do you portion off
01:01:26
the effort when you're making systems
01:01:28
right now in terms of energy efficiency,
01:01:31
the raw horsepower, and then the
01:01:33
transport layer? these seem to be the
01:01:35
three most important parts of what
01:01:38
you're doing. Um, correct me if I'm
01:01:40
wrong. And and then how how do you
01:01:42
allocate with engineers and your overall
01:01:45
team tackling those three major issues?
01:01:47
>> One way to think about it that that I
01:01:50
don't hear often enough is the way you
01:01:52
make a computer is you think about three
01:01:54
things, right? How fast you can do a
01:01:56
calculation.
01:01:57
>> Where you can store the result.
01:01:59
>> Mhm.
01:02:00
>> All right. Memory. and how fast you can
01:02:03
get the result to somebody who wants to
01:02:05
use it.
01:02:06
>> Transport,
01:02:06
>> not transport. These are the three
01:02:08
things that make a computer. And if you
01:02:10
do really fast calculations, but your
01:02:12
storage is is slow.
01:02:15
>> All right, bottleneck.
01:02:16
>> You're bottlenecked. If you can do fast
01:02:18
calculations, you can store it, but your
01:02:20
IO is slow, then you can't get it to the
01:02:22
user.
01:02:23
>> You are constantly thinking about as a
01:02:25
computer architect the balance
01:02:27
>> of these three dimensions, right? And so
01:02:30
you make a a jump in your in the the
01:02:32
performance of calculation. You got to
01:02:34
think about storage.
01:02:36
All right. Then you got to think I mean
01:02:37
it is a constant.
01:02:38
>> Are you thinking about those three
01:02:40
simultaneously or are there teams
01:02:42
grinding out each one of those
01:02:43
individual verticals? How do you
01:02:45
architecturally build a group of
01:02:48
engineers to do that? So you usually
01:02:50
your most senior architects, your your
01:02:53
CTO and your your technical leads are
01:02:57
thinking about that as the the the basis
01:03:00
of a design,
01:03:02
>> right? I mean it it it doesn't matter
01:03:03
how fast the car can go if it can't turn
01:03:06
you, right? It's not a good car except
01:03:09
maybe for drag racing, right? And so the
01:03:11
the designers are constantly thinking
01:03:13
about um where should we use power in
01:03:16
the design? What can we make the memory
01:03:18
faster? Can can we can we add memory?
01:03:21
What is the cost of adding memory versus
01:03:23
making it faster? The GPU, for example,
01:03:25
has a lot of capacity of memory, but
01:03:27
it's really slow.
01:03:29
All right? And that's a huge bottleneck
01:03:31
in inference. It's why they can't be
01:03:32
fast. It's why they just spent 2020 20
01:03:35
billion dollars buying Grock is because
01:03:37
they didn't have an answer for fast
01:03:39
inference. Fast inference needs fast
01:03:42
access to memory and the GPU doesn't
01:03:45
have it. So, these are things we're
01:03:47
constantly thinking of
01:03:48
>> and we're having a massive memory
01:03:50
shortage right now because of this. How
01:03:52
does that get resolved? Is that like
01:03:54
just a short-term bottleneck or is that
01:03:56
going to be a long-term problem?
01:03:57
>> I think what it's a a crazy problem. I
01:04:00
think everybody knew that the demand
01:04:02
would increase. Um, and this is true
01:04:04
among the the major memory makers. Um
01:04:07
people get a little scared and what
01:04:09
happens is they place a full year's
01:04:11
worth of demand and they get the wrong
01:04:13
answer back which is we don't we don't
01:04:16
exactly know h when you can have it. So
01:04:19
then their response is all right we'll
01:04:20
give you 18 months of demand. So
01:04:22
suddenly everybody went from giving six
01:04:24
months of demand to 18 months of demand.
01:04:26
>> Okay.
01:04:26
>> All right. Everybody all the supply
01:04:28
chain is confused. All right. We were
01:04:31
making the exact same amount of memory
01:04:32
we are now as we were four months ago.
01:04:35
All right. And what's happened is the
01:04:37
signal into the the makers has exploded
01:04:41
and it will take us about 18 months to
01:04:43
digest. The prices will stay high. Um
01:04:45
this is a known phenomenon in the memory
01:04:47
market. This happens every six or eight
01:04:49
years. Um what's different right now is
01:04:52
the the GPUs are using a huge amount of
01:04:54
HBM which is a flavor of DRAM and
01:04:56
they're chewing through that and that's
01:04:59
maybe leaving a little less for other
01:05:01
devices consumers. far along are the
01:05:03
Chinese in catching up to your company,
01:05:05
Nvidia, Grock, and how do you think
01:05:09
about the geopolitics of the AI race?
01:05:13
Like if they is there a scenario where
01:05:16
they win and we lose, we win, they lose,
01:05:18
or is that overblown in your mind?
01:05:21
>> I think the geopolitics are a real
01:05:23
issue.
01:05:24
>> Okay.
01:05:24
>> We are in we are well ahead in chipm.
01:05:28
>> Okay. within a few square miles of Santa
01:05:31
Clara. Uh you you had Intel, you had uh
01:05:35
AMD, you had Nvidia, you have our team,
01:05:38
you have ARM, you have one of ARM's
01:05:40
great teams. You you have maybe talent,
01:05:42
you have six of the world's great 10
01:05:44
chip teams.
01:05:46
>> Um I think the the way you get good at
01:05:48
building high-speed chips is to build
01:05:50
high-speed chips.
01:05:51
>> And when that that's really how you do
01:05:53
it and
01:05:54
>> you play the game, you get better at the
01:05:55
game.
01:05:55
>> That's right. Turns out you get better
01:05:56
at the game. and that that's been a a
01:05:58
weakness in the Chinese chipm ecosystem.
01:06:02
Now they're running hard and and they
01:06:04
know they're behind on that. I think on
01:06:06
the other side, I think they have pushed
01:06:08
ahead in the open model category.
01:06:11
>> Yes, the open source.
01:06:12
>> The open source model is an area where
01:06:14
they've pushed ahead. I think
01:06:16
>> because they're a top- down economy,
01:06:18
they were able to make decisions like
01:06:20
we're going to bring a huge amount of
01:06:21
power onto our grid. We're going to
01:06:22
modernize our grid. So they were able to
01:06:24
bring on huge amounts of power and
01:06:27
that's something that we're behind on. I
01:06:29
think it's unpleasant to to think of
01:06:32
them as as adversaries and we got to
01:06:34
figure that out together. Um I think the
01:06:36
world is a better place where we're not
01:06:38
adversaries but right now we are
01:06:40
>> and I I think certainly in an in an
01:06:42
industrial context we're we're
01:06:45
adversaries.
01:06:45
>> There's the industrial context and then
01:06:47
there's as we discussed what impact does
01:06:49
this actually have? What's downstream of
01:06:52
us winning? And it's every developer
01:06:54
being a 100x developer and then every
01:06:58
knowledge worker being a 100x knowledge
01:07:00
worker and every biotech innovation.
01:07:04
>> Systems that are recursive, right? That
01:07:06
build on themselves at rapid rates
01:07:09
>> have a huge winner take all feel, right?
01:07:11
Right. By getting ahead, you get further
01:07:13
ahead. Your iteration speed accelerates
01:07:16
and even small differences at the
01:07:19
beginning are magnified very quickly.
01:07:21
That's why this race is so important.
01:07:24
>> So, here we are. We're at Davos. It's a
01:07:28
lot of politicians.
01:07:30
My friend David Sax, co-host here on the
01:07:31
pod. He's our AIS are. And Trump,
01:07:35
whether you voted for him or not, is
01:07:37
very focused on this issue. Biden, their
01:07:40
team wasn't courting Silicon Valley. In
01:07:43
fact, they kind of looked at us as the
01:07:44
problem, demonized to a certain extent.
01:07:47
How how do you think objectively,
01:07:50
you know, independent of how you might
01:07:52
feel about ICE agents in our cities or
01:07:55
Greenland, etc. How do you think the
01:07:58
Trump administration is doing on their
01:08:00
AI policy and the support they're giving
01:08:02
the AI industry?
01:08:03
>> I think even a lot of fronts are doing
01:08:04
really well. I I I think
01:08:05
>> unpack it.
01:08:06
>> I think we we had made a mistake in the
01:08:08
previous administration keeping our
01:08:09
chips from our allies.
01:08:11
Let's keep China separate for a second,
01:08:13
but uh the UAE is clearly an ally.
01:08:16
absolutely
01:08:17
>> as an ally, right? Modern Arab nation,
01:08:19
been a source of peace, uh made peace
01:08:22
with Israel early on, huge Western
01:08:25
influence. Um and we kept chips from
01:08:27
them,
01:08:28
>> right? We like KSA, we like the Kingdom
01:08:30
of Saudi Arabia to move in the same
01:08:32
direction, kept chips from them,
01:08:34
>> right? We then made no sense. We we then
01:08:36
made a hierarchy
01:08:37
>> that that made the Danes feel second
01:08:39
rate, right? We said, "You are a number
01:08:41
two friend." Um bad idea. um we should
01:08:45
be empowering our allies, right? So
01:08:48
that's the first thing and and I don't
01:08:50
think the previous administration did a
01:08:52
good job. They didn't understand that at
01:08:53
all and Trump did a good job of that.
01:08:55
Not only do we want those institutions,
01:08:58
those nations and their institutions
01:09:00
using our technology, we want them
01:09:02
investing in the US and we under the
01:09:06
previous administration, we had a
01:09:07
cifhious organization in Treasury that
01:09:10
was was was difficult to work with and
01:09:13
that's all of those much improved and
01:09:15
>> they were unclear. They were not
01:09:17
communicative.
01:09:18
>> They they were impossible to deal with
01:09:19
>> impossible
01:09:20
>> impossible to deal with. This is super
01:09:23
important because
01:09:25
as David has said many times on this
01:09:27
program, hey, we want to be the
01:09:30
standard,
01:09:30
>> right?
01:09:31
>> And then all of that energy goes back
01:09:33
into our standard now
01:09:34
>> into our ecosystem in a development on
01:09:36
top of us into the recursive system we
01:09:38
just described.
01:09:39
>> Exactly. And if you look at Huawei,
01:09:41
right,
01:09:41
>> and what they did with 5G, their
01:09:42
networking up against Cisco and and you
01:09:44
know, our national champions, they they
01:09:46
ran the table in a lot of countries.
01:09:48
>> They clobbered us in Africa. They
01:09:50
clobbered us in the developing world.
01:09:52
They they absolutely ran the table.
01:09:54
>> Right. And now those places have
01:09:56
spyware.
01:09:57
>> That's exactly right.
01:09:59
>> And it's a real issue.
01:10:00
>> It's a real issue.
01:10:01
>> It's a real issue.
01:10:02
>> So I I think those were all areas where
01:10:06
this this administration did absolutely
01:10:08
the right thing.
01:10:09
>> Energy.
01:10:09
>> Uh energy. Another area. Um, I I think
01:10:13
one of the things that kills a company
01:10:15
like us is trying to grow extremely
01:10:17
quickly is having to deal with different
01:10:19
regulations in each of 14 localities
01:10:23
we're trying to put data centers.
01:10:24
>> Right.
01:10:24
>> Right. That that is brutal. Right. What
01:10:27
what you don't want when you're trying
01:10:28
to grow really quickly is to have 17
01:10:31
lawyers, each of whom is trying to
01:10:32
figure out the the local, right? And so
01:10:35
his effort to try and say, look, let's
01:10:37
get some reasonable laws across the
01:10:40
board. that's obviously smart and if we
01:10:42
could get some money to to to improve
01:10:44
the grid across the nation that also be
01:10:46
really helpful. All of those are
01:10:48
extremely positive. The work he's done
01:10:50
with the Department of Energy,
01:10:51
>> Chris, right?
01:10:52
>> Right. Under uh I think it's called uh
01:10:55
is it the Genesis program? I think it's
01:10:57
sort of the equivalent of a Manhattan
01:10:58
project for AI. Of course, we need this.
01:11:02
Of course, we need to be thinking among
01:11:03
our researchers, not how we can get a
01:11:05
little bit faster, 10 or 20, but what
01:11:07
can we use AI to to increase the rate of
01:11:09
research by 5x, by 10x, and how can we
01:11:12
get the the things that impede
01:11:14
government out of the way?
01:11:15
>> Yeah.
01:11:15
>> Right. So, those are good. China's a
01:11:18
really sticky problem. Um, and I I'm not
01:11:21
sure I agree with the the current push
01:11:23
to to allow the selling of of H100s
01:11:25
there, but it's reasonable to disagree
01:11:27
with me there. I I think it it is not it
01:11:30
is not clear-cut like some of the other
01:11:32
ones at all. It's a hard problem, and I
01:11:34
I think there going to be lots of
01:11:35
different views there.
01:11:36
>> Yeah. And I don't know if you've been
01:11:38
watching the news, but Canada just made
01:11:39
a strong alliance with China, announced,
01:11:41
I think yesterday or today when we're
01:11:43
taping this. And this is where, you
01:11:45
know, maybe the Trump administration can
01:11:47
improve is, you know, we we do need to
01:11:49
maintain this alliance with our
01:11:51
neighbors so that they don't they feel
01:11:54
like they can trust us. This is what
01:11:56
I've heard, you know, spending time in
01:11:57
Japan where the Japanese feel like maybe
01:12:01
we are not the most reliable partner.
01:12:04
Canada feels we're not the most reliable
01:12:05
partner because of the tariff issues,
01:12:07
military issues and maybe just the
01:12:10
constant changing of uh you know
01:12:13
consistency is really important in
01:12:15
policym and and they have to know
01:12:18
>> and I I think for a country like Canada
01:12:19
that has a huge amount of raw material
01:12:21
exports right they have wheat they have
01:12:23
lumber they have a huge amount of stuff
01:12:26
that that either we imported they got to
01:12:28
take elsewhere we we have to be aware of
01:12:30
of sort of the real politic of the
01:12:32
situation they have to sell their raw
01:12:34
material, right, which is a huge part of
01:12:36
their export. They have to sell it
01:12:37
somewhere and China is a big buyer,
01:12:40
right? And and so it's we have to go in
01:12:43
understanding that there are nations
01:12:46
which are proud and if you're constantly
01:12:49
attacking them and saying things to the
01:12:51
populace there, well then it gives the
01:12:53
leaders the ability to say, well, hey,
01:12:56
China is courting us and they're going
01:12:58
to invest and so why don't we build some
01:13:00
ports with them?
01:13:01
>> Yeah, ports. You should do a whole show
01:13:03
if you haven't already on the rise of of
01:13:06
Chinese ownership of major ports and
01:13:08
shipping. It is crazy when you look I
01:13:11
mean basically they own the world's
01:13:12
large shipping routes.
01:13:13
>> The belt and road strategy
01:13:14
>> and it's really
01:13:17
>> and then if you get out of sort of I
01:13:20
don't know I'm not here in in in
01:13:22
Switzerland very often but if you go
01:13:23
into many parts of the third world you
01:13:25
begin to see uh by was it BYD cars?
01:13:28
>> Yeah. They're gonna be shipping them to
01:13:30
Canada now,
01:13:31
>> all over the rest of the world. We We
01:13:33
don't see it, but it's unbelievable.
01:13:36
>> I was just in Mexico City with the wife
01:13:38
for a couple of days.
01:13:39
>> Isn't Mexico City fun?
01:13:40
>> I It's my first time there. I had a
01:13:42
delightful time. The food is
01:13:43
spectacular. I love it. Really fun.
01:13:47
Really fun, trendy, great place.
01:13:50
>> Yeah. Good vibes. And every guard of
01:13:52
BYD.
01:13:53
>> Yeah.
01:13:53
>> And we got to think about that. And
01:13:56
they're talk about national champions.
01:13:58
There's no way that the government isn't
01:14:00
subsidizing those by 30, 40, 50%. And I
01:14:04
think what their goal is to put the
01:14:06
Germans, you know, the English car
01:14:08
manufacturers, they've been at it for a
01:14:10
while, but the Germans are still making
01:14:11
pretty great cars. And if these BYDs,
01:14:14
which they're starting to get footholds
01:14:15
in Europe, the same thing will happen.
01:14:17
Like who's going to buy a 40, 50,
01:14:19
$60,000 Beamer, Volvo, Audi when you can
01:14:22
buy a 20, 30, $40,000 BYD? Yeah, they're
01:14:26
they're nice cars.
01:14:28
>> They're price dumping though, and that's
01:14:30
what tariffs are meant to protect
01:14:31
against.
01:14:32
>> They're they're subsidizing at the top
01:14:34
of the at the finished product,
01:14:36
>> right?
01:14:37
>> And that benefits the whole supply
01:14:39
chain, right? That's what they're trying
01:14:40
to do. They think about it as the
01:14:43
battery maker, the transmission maker,
01:14:45
all are benefiting while they subsidize
01:14:48
at the very top.
01:14:49
>> All right, let's end on employment.
01:14:51
Let's put the crystal ball out there.
01:14:53
Microsoft, Uber, Coinbase, Meta, Google.
01:14:57
Four years ago, five years ago, more
01:15:00
employees than they have now or they're
01:15:01
flat.
01:15:02
>> Yep.
01:15:02
>> Young people unemployment starting to
01:15:05
hit 10, 20% among some college age
01:15:07
demographics.
01:15:08
>> Yep.
01:15:10
>> You know, David Sax and I have this
01:15:11
debate all the time. Is it AI? Is it
01:15:13
entitled kids who don't have a work
01:15:15
ethic? Is it the overfunding and the
01:15:17
digestion or indigestion of tech
01:15:19
companies that hired two years out? It's
01:15:21
pretty clear to me watching startups who
01:15:23
are the most resourceful,
01:15:26
they're doing so much with AI. They are
01:15:28
AI first. They're building agents.
01:15:29
They're doing everything with AI.
01:15:31
>> There's no world in which we're not
01:15:32
going to have AI displacement.
01:15:34
>> Job displacement.
01:15:35
>> That's not why it's displaced now, but
01:15:38
100% it's coming.
01:15:39
>> Okay. So, when you look at it, you're in
01:15:42
the camp of it's coming, but it's not an
01:15:44
issue today. Define when it's coming.
01:15:46
what why it's not an issue today is when
01:15:48
I look at the uh
01:15:52
the people who have been let go
01:15:56
um in middle management in particular
01:15:59
>> okay be candid where you're on all in
01:16:00
you
01:16:01
>> no I mean this is middle management what
01:16:03
I think has happened is this is the
01:16:06
delayed impact of good SAS tools
01:16:09
>> ah that what's happened is your ability
01:16:13
to extend your reach as a leader and as
01:16:16
manager to stay a breast of what's
01:16:18
happening. Your scope is much much
01:16:20
bigger. And so the role of middle
01:16:22
management, which was frequently to move
01:16:24
information.
01:16:25
>> Yeah.
01:16:25
>> To manage small teams and move
01:16:27
information,
01:16:28
>> keep people on track.
01:16:29
>> That's right. That job h has shrunk in
01:16:33
value. I don't think it's yet AI.
01:16:36
>> I think halfway there.
01:16:37
>> That's right. I think AI is coming,
01:16:39
>> but I don't think that's what this is.
01:16:41
And what happened was there was this
01:16:43
ballooning of these jobs and you know
01:16:46
Mark Zuckerberg and and and Satcha they
01:16:49
look one day say competition is coming
01:16:52
it is much more intense. What are these
01:16:54
waves of people doing?
01:16:55
>> They're slowing us down let's be honest.
01:16:57
>> That's right. And so they they're
01:16:58
flattening their organizations as well.
01:17:01
So it's not just they're moving people
01:17:02
out but they're changing the the shape
01:17:05
of the organization which is why I don't
01:17:07
think it's AI yet.
01:17:08
>> Yeah. I think what we're going to see
01:17:09
down the road is whole categories that
01:17:12
that are vastly more efficient and
01:17:15
therefore need less people.
01:17:16
>> It's uh pretty clear and it's almost
01:17:20
rest in peace Scott Adams, creator of
01:17:22
Dilbert, but
01:17:22
>> huge fan.
01:17:24
>> Yeah, he just passed away this week. I
01:17:25
>> I saw that. Huge fan.
01:17:26
>> What a giant.
01:17:27
>> What a giant of of ridiculing uh
01:17:30
corporate America. That exact
01:17:33
layer is gone.
01:17:34
>> Corporate America. It's actually great
01:17:36
that Scott got to see it.
01:17:38
>> Yes.
01:17:38
>> Happen. Yeah.
01:17:39
>> Towards the tail end. And uh we didn't
01:17:41
get to mention it on a previous episode,
01:17:42
but rest in peace, Scott Adams. I think
01:17:44
it's a good place for us to uh end
01:17:45
there, Andrew. I know you got a lot to
01:17:46
do here. Enjoy your time at Davos.
01:17:49
>> Thank you.
01:17:50
>> Yeah. If you see any of the Germans, ask
01:17:51
them why they turned off their nukes.
01:17:53
>> All right.
01:17:53
>> Yeah, that's right there to joke to
01:17:56
them. How's Greta Thurber? How is your
01:17:58
secretary of energy doing, Greta
01:18:00
Thurberg? Like, what are you doing? They
01:18:02
turned off three of their six nuclear
01:18:04
reactors. I know. So they they decided
01:18:06
instead to import natural gas from
01:18:08
Russia.
01:18:08
>> Where did they get it from?
01:18:08
>> Right. From Russia.
01:18:09
>> Oh, from Russia.
01:18:10
>> Yeah. Bad. Depended on Russia. Really
01:18:12
not smart.
01:18:13
>> Yeah. Not smart.
01:18:14
>> And you know what? All because of Davos.
01:18:16
I blame the WF and Davos. They literally
01:18:18
got so caught up in virtue signaling
01:18:20
about the environment. They never just
01:18:22
looked from first principles at how safe
01:18:24
nuclear is compared to burning fossil
01:18:27
fuels.
01:18:28
>> Nuclear is is is safe and we can make it
01:18:30
safer. We got to put the time and effort
01:18:32
in. I was just in Japan last week.
01:18:34
>> They're putting in new nuclear reactors
01:18:36
and they just got over a Fukushima and
01:18:38
they realized, "Oh, we made some
01:18:39
mistakes putting it below sea level.
01:18:41
>> We're not going to make those mistakes
01:18:42
again. Nuclear is obviously the way to
01:18:44
go."
01:18:44
>> Let's get better at it.
01:18:45
>> Hey, let's get better at it. That was
01:18:46
awesome, dude.
01:18:47
>> Thanks for all the time and for a great
01:18:49
discussion. You rocked it. All right,
01:18:51
everybody. Welcome back. We're grifting.
01:18:53
I mean, interviewing the top CEOs here
01:18:56
at Davos, the World Economic Forum. Uh
01:18:58
this is our first time here and we're
01:19:00
having a great time. Tons of CEOs. We've
01:19:03
had hundreds of interview requests.
01:19:05
We're going to try to do about a half
01:19:06
dozen of them. A friend of the pod, Jake
01:19:09
Lucerarian, is here. Uh you've been on
01:19:12
the pod before, both this week in
01:19:13
Startups and Allin. Uh you of course the
01:19:15
CEO and co-founder of Gecko Robotics.
01:19:18
You've been at it for almost a decade
01:19:20
now. Yeah.
01:19:20
>> You build robots, as people who have
01:19:22
seen the pod before know, inspect. These
01:19:25
are purpose-built robots that will
01:19:26
inspect ships, bridges, whatever it
01:19:29
happens to be. And you started this long
01:19:31
before Chad GPT and this recent AI
01:19:33
revolution. I'm curious, these robots
01:19:35
which are very purpose-built, you know,
01:19:37
I think very straightforward.
01:19:40
>> Have you started to put AI into them
01:19:41
yet? Because I I was just curious
01:19:44
thinking about your previous
01:19:45
presentations. It was pretty
01:19:47
straightforward, right? Like we know the
01:19:48
bridge, inspect the bridge, but
01:19:50
>> now can it do things and start thinking
01:19:52
on its own and maybe be more adaptable
01:19:54
because of AI?
01:19:55
>> Yeah, good question. Well, I changed my
01:19:56
title now. It's now chief uh grifting
01:19:58
officer.
01:19:58
>> Oh, chief drifting officer.
01:19:59
>> Thank you. Especially when I'm in Davos.
01:20:01
This is my title.
01:20:01
>> You've been here a couple of times.
01:20:03
>> Yeah, exactly.
01:20:04
>> Did you catch the tail end of the DEI?
01:20:07
And
01:20:08
>> I came I came right at the like the
01:20:10
heart of it. Yeah. And so it was it was
01:20:12
you had to learn a different language
01:20:13
actually.
01:20:14
>> Really? Yes. Did they check your fluency
01:20:15
in ESGdei buzzwords?
01:20:17
>> They did. They did. Well,
01:20:18
>> no, but everything was super precious
01:20:19
and now I guess since Trump is here,
01:20:22
>> it's kind of brass tax like doing
01:20:24
business negotiating and like less of
01:20:26
this performative stuff.
01:20:27
>> It's a lot actually performance tonight.
01:20:29
Um, but there's a lot there's a lot more
01:20:31
focus on okay, let's get down to the to
01:20:34
the brass tax. We hear a lot of CEOs
01:20:36
talking about AI and but actually a lot
01:20:38
of the conversations I'm having in the
01:20:39
Congress already is just about okay,
01:20:41
like where's the ROI from all the AI?
01:20:44
business.
01:20:44
>> It's actually business. It's actually
01:20:46
trying to get to the to the first
01:20:47
principles to the roots of okay, how do
01:20:49
you actually get artificial intelligence
01:20:51
to to deliver on the promise and funny
01:20:53
enough a lot of it comes down to this
01:20:54
this really interesting uh gap that's
01:20:57
exists in AI which is like all the
01:20:58
information and data set that you need
01:21:00
to actually turn all this into actual uh
01:21:03
return on investment and productivity
01:21:05
gains especially for these like large
01:21:07
infrastructure large asset owners uh
01:21:09
like the energy or mining or
01:21:10
manufacturing companies of the world. So
01:21:12
that's a big focus and that's what
01:21:14
>> it really seems to be turning into a
01:21:15
business conference. I was astounded by
01:21:18
the amount of inbound I had that was
01:21:20
just pure business capitalism building
01:21:23
products and services to make life
01:21:25
better. Also the world has changed a lot
01:21:28
since you started the firm.
01:21:30
>> Um we've got a new sort of military 2.0
01:21:34
know thing happening and I think a lot
01:21:35
of your customer base moved from just
01:21:38
maintenance of bridges and tunnels and
01:21:40
infrastructures to military. Tell us
01:21:41
about that.
01:21:42
>> Yeah, that's exactly. Yeah, that's
01:21:43
right. We we do about 30% of our
01:21:45
business is defense. So, we we work or
01:21:47
department of war, I guess I should say.
01:21:48
So a lot of it's focused on how do you
01:21:51
actually use technology to you know
01:21:53
fight against the speeds of development
01:21:55
of of countries like China for example
01:21:56
in terms of manufacturing speeds and a
01:21:58
big part of that is actually
01:22:00
understanding the quality and the and
01:22:01
the welds um the putting together the
01:22:03
actual welds put pieces together. This
01:22:05
is actually a huge bottleneck for the
01:22:07
US. We have you know these manufacturing
01:22:08
and forges that are 100 years old doing
01:22:10
things 100 year old um you know today
01:22:11
like they did you know back then. So,
01:22:13
the technology that we're using to
01:22:14
deploy to help manufacture um certain,
01:22:17
you know, components of uh uh a
01:22:19
submarine or or be able to help expedite
01:22:22
how fast a destroyer is in turn around
01:22:23
to get out and patrolling borders and
01:22:25
deterring conflict. Um these are sort of
01:22:27
the things that our robots are using to
01:22:29
help to speed up decision-m process and
01:22:31
make sure you're accurate. Um and so in
01:22:33
some cases that Admiral Houston has
01:22:35
talked about 90% improvements to speed
01:22:37
up manufacturing using the technology
01:22:38
that Gecko builds and you know, you're
01:22:40
seeing companies like Ander now um you
01:22:41
know, working with us. Palmer lucky.
01:22:43
>> Palmer from lucky. Yep. Of course. And
01:22:44
>> that's his helicopter up there. He might
01:22:46
drop a bomb any minute.
01:22:47
>> Yeah. I think he's doing a speech pretty
01:22:49
soon. Um and so um so it's just amazing
01:22:51
to see the adoption. But on the energy
01:22:53
side, right? Like that's been the
01:22:55
biggest the biggest growth areas for our
01:22:56
company. It's just been these large
01:22:59
energy companies and power companies.
01:23:01
They're trying to figure out, okay, all
01:23:02
of these um all these hyperscalers are
01:23:05
trying to figure out how to get
01:23:06
infrastructure. They're focusing on
01:23:07
capex a lot, right? Well, what if we
01:23:09
started to play a game where, you know,
01:23:10
we have access to these problems to
01:23:12
these really um, you know, these GDP
01:23:14
driver companies. What if we actually
01:23:16
took an AI native or in our case what
01:23:18
we've seen a lot of companies be is like
01:23:19
how to be robot native first
01:23:22
>> um to support the AI initiatives um by
01:23:23
injecting and taking a very aggressive
01:23:25
approach and how to implement and and
01:23:27
put robotics to use to help build up the
01:23:29
the data infrastructure to then layer on
01:23:31
AI models. And so that's at the heart of
01:23:33
what Gecko does. That's why I started
01:23:34
this company 13 years ago with this
01:23:36
premise of data matters as it relates to
01:23:38
being able to uh to have all the games.
01:23:40
>> For people who don't know, the robots
01:23:42
have sensors in them, different arrays
01:23:44
that can inspect metal, whatever the
01:23:47
fabrication is, and go right to the
01:23:49
seams of a submarine and make sure this
01:23:52
is all been done perfectly
01:23:54
>> and measure it perfectly.
01:23:55
>> That's right. We build the robots and
01:23:57
the sensors that go around and look at
01:23:58
diagnosing the health of the built
01:24:00
world. So that means just like
01:24:01
understanding and getting the largest
01:24:02
inventory and database of information
01:24:03
about the health of build structures,
01:24:05
bridge, dams, submarine, whatever it is.
01:24:07
Um now in along that journey, you're
01:24:09
able to figure out that if you
01:24:10
centralize all that information and data
01:24:11
and then layer on top of it operational
01:24:13
data which exists, you know, for the
01:24:14
most part is a decent like
01:24:15
infrastructure of sensor data at these
01:24:18
companies. Well, wow, you get to make
01:24:19
some pretty interesting decisions
01:24:20
because you can figure out how to extend
01:24:21
the useful life of an asset. And if I
01:24:23
push an asset harder, can I produce
01:24:25
more? like wow my focus is how do I help
01:24:27
to create cleaner as well as um more
01:24:30
barrels per day and at lower costs. I
01:24:32
use the word cleaner.
01:24:33
>> So if you have a refinery or you have a
01:24:35
nuclear power plant, you inspect it.
01:24:37
>> Robots should be dedicated to figuring
01:24:39
out how to solve for the business
01:24:41
problem. The b what is the fundamental
01:24:43
business problem that the customer is
01:24:44
trying to to solve for? If it's making a
01:24:46
barrel or making a kilowatt or getting a
01:24:48
ship out of dry dock faster, that is our
01:24:49
initiative and our goal as a company to
01:24:51
build robotic solutions towards that.
01:24:53
Now, I haven't built we haven't gotten
01:24:54
into building the humanoids of playing
01:24:56
the humanoid game. And that's actually
01:24:57
when I was on your podcast, actually the
01:24:58
summit, I talked about we're going to be
01:25:00
the biggest um purchasers of the Optimus
01:25:02
robot is because the real question is
01:25:04
how do you actually employ robots? How
01:25:05
do you get robots to return ROI?
01:25:08
Because, you know, folding laundry and
01:25:09
cleaning dishes is not a high ROI use
01:25:11
case. It's going to be the $20 an hour.
01:25:14
>> You're not going to pay 40,000, you
01:25:15
know, for or 20,000 whatever it is. But,
01:25:18
you know, US has to be the best in the
01:25:19
world at figuring out how to use robots
01:25:21
to make um unfair advantages, you know,
01:25:24
with these companies, whether it's oil
01:25:25
and gas, whether it's
01:25:26
>> power need somebody between Tesla or
01:25:29
Figure or Boston Dynamics, those robots
01:25:32
are going to be sold. there's going to
01:25:34
be need to be an application layer and
01:25:36
operational excellence in the field will
01:25:39
be that that's exactly what we are with
01:25:40
a nervous system to pull all this
01:25:41
information from robots together and
01:25:43
then you can build and use um AI models
01:25:45
on top of that to use the information
01:25:46
data to then also begin to take actions
01:25:48
back into the real world. So you
01:25:50
actually see it find not just right now
01:25:53
you find problems or monitor situations
01:25:56
and confirm things are being built
01:25:57
properly that there's no potential
01:26:00
problems with this nuclear power plant
01:26:02
or this ship but down the road do you
01:26:04
see yourself actually taking actions to
01:26:07
build andor repair? Yeah that's exactly
01:26:10
the road map for us but first you have
01:26:11
to figure out like what what is the
01:26:13
state of the built world? What is the
01:26:14
state of the health? Um what sorts of
01:26:16
actions should I take when it comes to
01:26:17
repair? What sorts of automated welding
01:26:19
solutions, for example, are the right
01:26:21
ones and which ones could use
01:26:22
information about the how well that weld
01:26:24
was done is a feedback you to create a
01:26:26
foundation model for welding, being the
01:26:28
best in the world at welding. And so the
01:26:29
archite
01:26:32
>> we're going to Yeah, we're going to be
01:26:33
the the company that builds robots um to
01:26:35
both identify and then solve for the
01:26:37
most important and highest ROI problems
01:26:39
for the customers, whether they're
01:26:40
manufacturing new assets or they're or
01:26:41
they're trying to operate and maintain
01:26:43
existing ones. Funny enough, like I've
01:26:45
been talking a lot about like how do I
01:26:46
reduce hazardous work hours for humans?
01:26:48
How do I extend the use life of assets
01:26:49
for infrastructure? How do I, you know,
01:26:51
increase the capacity and production uh
01:26:53
and uh and prevent, you know, catch
01:26:55
failures from assets. Funny enough,
01:26:56
these are all very like easy to
01:26:57
underwrite problems and it's something
01:26:59
that you don't hear roboticists or AI
01:27:01
founders talk a lot about, but like
01:27:03
that's my bread and butter. That's the
01:27:04
world I live in and I wear the steel toe
01:27:05
boots to be able to understand the
01:27:06
problems. You're going to need humans in
01:27:08
the loop for some time to come and there
01:27:10
are there's going to be plenty of work
01:27:12
for welders, but there might also be
01:27:15
incremental jobs created because
01:27:17
>> yeah,
01:27:18
>> and you'll have one welder maybe
01:27:19
supervising 10 of these robots. Is that
01:27:21
what you think is going to happen?
01:27:22
>> That's what's going to happen. You want
01:27:23
to be able to get the experience and the
01:27:25
subject under expertise to make sure the
01:27:26
robot is actually understanding like
01:27:28
what what are the sorts of ramifications
01:27:30
if I do this action versus that action.
01:27:31
be able to, you know, you you want also
01:27:33
be able to what we don't, you know, need
01:27:35
to understand too is there going to be a
01:27:36
lot of tele operations, you know, with
01:27:38
mobile robots in particular. And so
01:27:39
you're going to have humans in the loop
01:27:40
just they might not be in the field as
01:27:42
much. They might be more, you know, in a
01:27:43
in an AC, you know, uh, building, you
01:27:45
know, being able to operate and and
01:27:47
build build information and data to
01:27:48
train the foundation model. If you think
01:27:50
how risky some of these jobs are, uh it
01:27:52
might be nice to not have a human
01:27:54
risking their life to maintain, you
01:27:56
know, this part of the bridge,
01:27:58
>> you know, as brave and amazing it is
01:28:00
they're doing that work. And we
01:28:02
obviously appreciate appreciate that
01:28:04
over the centuries.
01:28:05
>> Yeah.
01:28:05
>> It might be nice to actually take the
01:28:07
human out of the deep sea welding and
01:28:09
out of the bridge climbing business.
01:28:11
>> That's exactly right. Well, I mean, the
01:28:12
story of Gecko has been a story of
01:28:13
building robots to help to reduce the
01:28:15
barrier to entry of these like jobs that
01:28:17
sometimes take 10,000 hours to be great
01:28:19
at and actually making it something, you
01:28:21
know, you can attain within a few months
01:28:23
of being able to, you know, use the
01:28:24
technology in these fields. And so, you
01:28:26
know, whether it's manufacturing
01:28:27
different parts and inspecting the
01:28:28
quality of those parts or it's actually
01:28:30
gathering information, data set,
01:28:31
understanding what kind of decisions to
01:28:32
make. My goodness, like you have a
01:28:34
shortage of welders, you have a shortage
01:28:35
of inspectors, you have a shortage of
01:28:36
all these trades and you have to be able
01:28:38
to augment to take a Home Depot employee
01:28:39
in a couple months, make them, you know,
01:28:41
able to create a, you know, make
01:28:42
100,000, $150,000, you know, running
01:28:44
your robot, doing it safely. And so, you
01:28:46
know, that's an exciting, bright future.
01:28:48
I think the the key unlock for us, you
01:28:50
know, in the robotics community is just
01:28:51
like you got to get your robotics into
01:28:53
the field. Uh, you got to um you got to
01:28:55
fail fast and and also rapidly prototype
01:28:57
really quickly. And then manufacturing
01:28:58
them is is the big is the big issue. And
01:29:00
so you as we focus on these sorts of
01:29:02
problems because you know in this
01:29:04
journey in over the next 5 years you
01:29:06
know to be that company that's the best
01:29:08
in the world of taking robots and making
01:29:09
ROI from them you know we uh that's
01:29:11
that's what we're focused on in terms of
01:29:13
setting ourselves up to be the world
01:29:14
dominant there.
01:29:14
>> And you uh wrote an editorial on the way
01:29:16
in here and dropped it. What was your
01:29:18
take in the editorial?
01:29:19
>> It was basically on the concept that
01:29:20
we're talking about here which is like
01:29:22
you know I'm a I'm a I'm a I live in
01:29:23
Pittsburgh. You know at some point
01:29:24
Pittsburgh had like more millionaires in
01:29:26
1930 than New York. It made 70% of the
01:29:29
world's steel. Um, and it's, you know,
01:29:30
the economy changed a bunch and and that
01:29:32
doesn't happen anymore in in Pittsburgh
01:29:34
and and so but but like the insights and
01:29:37
the the companies there like they're
01:29:38
still the backbone of our of our economy
01:29:40
and and I was just inspired by, you
01:29:42
know, in that time frame, the industrial
01:29:44
revolution, how steel, you know, was
01:29:46
invented and then manufactured and then
01:29:48
distributed to help create all the
01:29:49
infrastructure that we rely on and, you
01:29:51
know, that that that fuels the energy or
01:29:52
fuels the manufacturing sectors. you
01:29:54
know, it was the it was the
01:29:56
infrastructure that you needed to be
01:29:57
able to have all these big gains um that
01:29:59
came from the industrial revolution.
01:30:00
Same thing is is what I was talking
01:30:02
about in the editorial about uh is what
01:30:04
we're doing with robotics collecting
01:30:06
information and data sets um to help
01:30:08
support um and create uh the avenue uh
01:30:11
create the infrastructure for AI models
01:30:12
to actually be able to you know return
01:30:14
the kinds of returns that we're all
01:30:16
betting on. So, you know, you it's just
01:30:18
important people to understand like that
01:30:20
the the robotics is the is the you know,
01:30:21
it's almost the it's almost the the
01:30:23
foundation to be able to get the massive
01:30:25
returns in the sectors that were that
01:30:26
are mostly all here. Great. Yeah. Being
01:30:29
able to eliminate some Dilbert level
01:30:32
middle managers who aren't adding value
01:30:35
>> with, you know, some automation. Okay,
01:30:37
fine. Uh but we really need to get out
01:30:39
there in the real world to get that
01:30:40
serious ROI, whether it's a robo taxi or
01:30:43
self. And and I think that the risk you
01:30:45
have like with like I think this the
01:30:47
forum is like it's like you're saying
01:30:48
it's changing right there's like a
01:30:49
there's a ecosystem in a bubble when you
01:30:51
live in a certain place talking a
01:30:52
certain way in Silicon Valley we we only
01:30:54
exist in the world of of the internet.
01:30:56
We we don't build the kind of
01:30:57
technologies that started in Silicon
01:30:59
Valley. And so this world of energy of
01:31:01
metal manufacturing of of mining all
01:31:03
these like sectors defense like they're
01:31:05
they're just not and when I was starting
01:31:07
the company they were like taboo to talk
01:31:08
about. So you just don't think about the
01:31:10
kind of you know applications and things
01:31:12
you
01:31:12
>> in some ways we ran out of things to
01:31:14
solve for. I mean like what's the next
01:31:16
when I would be pitched 10 years ago on
01:31:18
SAS software it was like okay great and
01:31:20
then it was okay this is the 50th SAS
01:31:23
software company in this vertical okay
01:31:24
this is the 15th in this vertical we're
01:31:26
kind of running out of
01:31:27
>> what's crazy to me
01:31:28
>> those verticals to go after.
01:31:30
>> You're exactly right and this is why
01:31:31
like you think of like an incredible
01:31:32
invention like a humanoid robot and the
01:31:34
first demo that me and you saw right was
01:31:36
a folding laundry. Oh my goodness. like
01:31:38
the the like that that was the the thing
01:31:40
that you know I do when I go home. So
01:31:41
that's what a robot should do. There's
01:31:43
so many other important like really
01:31:44
important applications to use this
01:31:46
technology for and I live in Pittsburgh,
01:31:48
right? So I get to see a completely
01:31:49
different world with a completely
01:31:50
different little bubble. Um, and then
01:31:52
you know going obviously and talking to
01:31:53
to CEOs of energy companies all day like
01:31:55
it changes your world and so like it
01:31:58
really is an advantage and I encourage a
01:32:00
bunch of the startups that you're
01:32:01
talking to you know to to focus on like
01:32:03
where the eyes are not where the
01:32:04
conversations are not go out of the
01:32:05
ecosystem and find the really important
01:32:07
problem to solve. If you think about it,
01:32:09
we had Boston Dynamics doing backflips
01:32:11
with these robots, you know, a decade
01:32:14
ago, but they they didn't have an LLM
01:32:16
behind them or a vision model or a world
01:32:20
model yet.
01:32:21
>> Yeah. And now when they have it, you'll
01:32:24
be able to, I think, tell it, "Hey, I
01:32:25
want to lay some bricks
01:32:27
>> and it'll just go out to the web and
01:32:29
find all the brick laying YouTube videos
01:32:32
and the history of brick laying
01:32:35
and every manual on brick laying and
01:32:37
every skew of every device ever used for
01:32:40
brick laying and it's going to know
01:32:43
>> how to do it
01:32:44
>> without ever having to be trained or is
01:32:47
that the the you know, I'm just giving a
01:32:49
very silly example, but
01:32:51
>> when would be able to just tell Optimus,
01:32:53
"We need you to lay some bricks." And it
01:32:54
goes, "I know kung fu." Boom. It just
01:32:56
does it.
01:32:57
>> Yeah, that's right. I think that the
01:32:59
>> How soon?
01:32:59
>> How soon? Uh, it's I don't think that's
01:33:02
going to be as far out. I think it's
01:33:03
like I I I take maybe take like more
01:33:05
like the three-year kind of time frame
01:33:06
for those kind of things. I think you
01:33:08
can kind of see this these like um you
01:33:10
know, these big bets like SoftBank um
01:33:12
and Nvidia I think just put a billion
01:33:13
dollars into um Skilled AI, which is
01:33:15
creating the brain for robots. Um
01:33:18
actually a $14 billion valuation. um the
01:33:20
founder of which is in Pittsburgh by the
01:33:22
way, Deepo. Um but I think the I mean
01:33:24
the big the big problem I think with
01:33:25
like that um extrapolation is in the
01:33:29
world of the world that I live in every
01:33:30
day whether you know the energy the the
01:33:32
the the defense etc etc. Um we don't
01:33:35
have those videos. There is not a corpus
01:33:37
of information in data set and so I'm
01:33:39
focused on that. I'm focused on
01:33:40
>> how do you get that data? You put GoPro
01:33:43
cameras and sensors on people's arms
01:33:46
like I saw
01:33:47
>> totally training. Is that how the
01:33:49
training will be done or where you're
01:33:52
like modeling and actually watching a
01:33:53
human do it and then having the robot
01:33:56
analyze it or
01:33:57
>> Yeah, it's about uh we we think about
01:33:59
like you know there's not many customers
01:34:01
that are going to pay for that cuz the
01:34:02
ROI is just like not not clear, not
01:34:05
there like no no big like energy or you
01:34:07
know manufacturing company's going to be
01:34:08
like yeah let's do that and I'll pay you
01:34:09
like 10 million bucks a year to do that.
01:34:11
And so like we're collecting it by
01:34:13
solving for you know important problems
01:34:15
on critical infrastructure and assets of
01:34:16
which like you know we're we're walking
01:34:18
around these like Manhattan size
01:34:19
refineries all the time. And so there's
01:34:21
information data sets that were like
01:34:22
building
01:34:22
>> and the refinery inspection is done by a
01:34:25
human today. Yeah.
01:34:26
>> They take a bunch of pictures. They they
01:34:27
use a bunch of sensors and now the robot
01:34:29
>> 100t up in the air on a on a rope
01:34:31
collecting data by hand. It's, you know,
01:34:33
so if you use a robot that has a bunch
01:34:35
more sensors to fuse together, it begins
01:34:36
to create a world that doesn't exist on
01:34:38
the internet, which gives Gecko a very
01:34:40
big advantage.
01:34:41
>> That's a that's a lot of world building
01:34:43
you're doing.
01:34:44
>> That's exactly right.
01:34:45
>> Continued success and we'll see you next
01:34:47
time on the AllIn interview program.
01:34:49
Bye-bye.
01:34:50
>> Good job.
01:34:50
>> Nice.
01:34:50
>> Awesome.
01:34:51
>> That was fun.
01:34:51
>> Oh man, cold out here, huh?
01:35:09
I'm going all in.
